It was -40°F out. 30 mph winds scoured exposed skin. 100 miles into my great winter adventure, on pace, on plan, executing months of preparation. I tested gear in subzero conditions. I dialed in nutrition. I studied the course. Then my rear tire goes flat.

The repair in the waning afternoon light eats vital time. My fingers and face, already numb from the cold, edge toward frostbite. By the time I reach the valley checkpoint, half a day later, race officials report carnage on the trail around me. Frostbite, overflow, open water, whiteouts. The psychology lesson is clear and harsh. Perfect preparation and relentless execution do not guarantee success. You can do everything right and still lose. Mental toughness alone is insufficient. Alignment matters. Luck decides.

Mental Toughness Without Alignment Accelerates Failure

The self-help industry sells mental toughness as a universal solvent, claiming perseverance alone conquers any obstacle. Psychological research tells a different story.

Mental toughness alone does not create success. Grind amplifies whatever direction you point. Point in the wrong direction, and more effort accelerates failure.

Angela Duckworth’s 2007 research on grit, published in the Journal of Personality and Social Psychology, shows perseverance and passion for long-term goals predict achievement. Her West Point study found grit the best predictor of completing Beast Barracks, outperforming SAT scores, high school GPA, and physical aptitude tests. Among National Spelling Bee finalists, grittiest contestants advanced further because they studied longer and with greater consistency, not because they were smarter.

But grit has limits. Duckworth herself admitted in a 2016 interview that many factors contribute to success beyond individual perserverance. She noted that money, social barriers, and other external constraints affect outcomes, meaning grit alone does not guarantee achievement. Did she grind to get her research results out there, did she have the resources to get published, did she know the right people, or did she get a lucky break?

Even worse, the sunk cost fallacy traps the grittiest individuals in failing endeavors. Having invested the most effort, they cannot admit it was misplaced. They keep grinding in the wrong direction because quitting feels like betraying their past self.

The formula requires more than effort. Apply effort in the right direction, toward solvable problems. Grinding only works when aligned with where you are actually headed.

Luck Decides

If alignment determines direction, luck determines outcome. That presents a key question. Can you influence luck?

Richard Wiseman’s decade-long study of 400 volunteers answers definitively: yes, but indirectly. His 2003 book The Luck Factor identified four principles that distinguish self-described lucky people from unlucky ones:

1. Opportunity creation. Lucky people actively create, notice, and act upon chance opportunities that unlucky people miss.
2. Intuitive decision-making. Lucky people trust their gut feelings to make better decisions.
3. Self-fulfilling prophecies. Lucky people create positive expectations that unconsciously influence their behavior.
4. Resilience. Lucky people adopt attitudes that transform bad luck into good.

The key nuance is that lucky people practice behaviors that maximize the probability of good luck, not guarantees. Wiseman’s lucky subjects experienced misfortune at statistically similar rates to unlucky subjects. The difference lay in how they positioned themselves relative to opportunity.

Consistent effort over time puts you in more situations where luck operates. The more you grind, the more lotteries you enter. But this only works when grinding aligns with meaningful objectives.

Consider SonReal. The Canadian rapper’s 2026 song-a-week proves weekly creation works. Releasing an unreleased track every week created weekly opportunities for discovery. But the alignment, creating music he genuinely believed in, working with his friends, and connecting directly with fans, made those opportunities valuable. The grind created quantity. The alignment ensured quality.

Extreme Sports Prove the Grind-Luck Equation

Extreme endurance sports provide a lab for studying the grind-luck equation. The competitors are elite, and all are well prepared. So the playing field levels as much as nature allows. Luck’s variance remains.

Lance Mackey’s career epitomizes the principle. He won the Iditarod Trail Sled Dog Race four consecutive times from 2007-2010, an unprecedented achievement that demonstrates exceptional capability. Yet in other years, he scratched due to health complications from throat cancer treatment and extreme trail conditions. Same athlete. Same effort. Same preparation. Weather changed. Snow conditions shifted. Dog health faltered. Luck turned. Effort alone does not explain the outcome.

Henry Worsley’s 2016 solo Antarctic crossing attempt offers a starker example. An experienced polar explorer with successful prior expeditions, Worsley prepared meticulously. But 900 miles into his solo crossing, his health began to fail. Luck decided the outcome, despite the processes. Mental toughness cannot repair failing health in Antarctica.

Reinhold Messner and Peter Habeler’s 1978 oxygen-free Everest ascent seemed scientifically impossible at the time. Their success proved exceptional preparation and mental toughness. But Messner also witnessed numerous friends die on Everest and other peaks from random avalanches, sudden storms, and falls. Same skill, same preparation, but timing shifted and conditions changed so the luck differed.

The pattern across these cases is effort creates possibility.

The Silent Graveyard of Grinders

If luck plays such a significant role, why do most people still believe effort alone determines success?

Survivorship bias. During World War II, analysts studied returning aircraft only, missing where enemy fire hit the missing ones. I’ll let you imagine the overused image of the fighter plane with holes in it that all data scientists know and love.

We study the winners and naturally assume their characteristics caused their success. We read their biographies and extract lessons about perseverance, passion, and hard work. We ignore the equally talented, equally hardworking individuals who failed due to circumstances beyond their control. The silent graveyard contains no memoirs.

In The Success Equation, finance theorist Michael Mauboussin distinguishes between skill-dominant domains and luck-dominant domains. In skill-dominant domains like chess, the best player wins consistently. In luck-dominant domains like entrepreneurship, investing, or endurance racing, the best performer wins only slightly more often than average and sometimes loses.

Mauboussin’s framework explains why the scratch rate is so high in some winter fatbike ultras. Half the field scratches due to equipment failures that cannot be prevented, weather events that cannot be predicted, or navigation errors in whiteout conditions that cannot be avoided. Perseverance alone cannot overcome these factors.

Effort still matters though. In luck-dominant domains, skill and preparation increase the probability of success. But the distinction between probability and certainty is key. The self-help narrative confuses increased probability with guaranteed outcome.

The Principles

Extreme endurance cases validate the psychology literature revealing three principles:

• First, grinding is necessary to keep things moving no matter where you are headed. Mental toughness can get you through a lot of shit.

• Second, grinding is only good when aligned with where you are actually headed. Misaligned grind is counterproductive.

• Last, you cannot control luck, yet you influence your exposure to it. Reacting increases your luck.

A few key questions remain. When do you keep pushing? When do you quit? When do you try something different?

In Part Two next week, I frame success as a problem of alignment. Back-of-the-envelope calculations quantify the trade-offs between grinding on your current path versus reassessing your direction. I discuss how built-in realignment in Stochastic Gradient Descent algorithms prevents misalignment from trapping deterministic approaches. And I apply these ideas to real-world exploration strategies in geology and other fields.

The psychology explains luck’s role and how to position yourself relative to it. The mathematics defines actions to take.

The air hung heavy before the rain. Thick with the scent of livestock and fresh saw dust swirling in the dim light filtering through the large tent covering the sale. Above, the sky pressed down, gray and cool, the kind of late summer day that promises rain but holds it just out of reach. In front, the auctioneer’s chant echoed off the packed bleachers, a rhythmic staccato that rose and fell like the breath of the fair itself. The low of cattle and the bleat of goats and lambs mixed with the clink of gate latches and the soft thud of hooves on wood chips. Neighbors leaned closer to hear each other, talking in voices pitched to carry over the din, their words punctuated by the occasional silence of an animal just sold.

Here community is alive in the shared air and shared purpose. The middle-schooler down the road raised that red steer in lot 4, fed it, walked it, knew its temper. Now it stands square and calm while bids fly and heads nod. The FFA jacket he wears is clean, but the early mornings, late nights, sweat and tears that went into raising that animal cling to it. The same kind of work that built this place and the people in it. When the gavel falls, it delivers recognition and belonging.

The kid feels it in their bones. That steer represents the weight of daily chores before school, the pelting spring rains on early feedings, the patient teaching of an animal to lead. The crowd sees the result. The kid knows the cost. And that shared knowledge binds the room.

Honest Work

Hard work done in common view earns its place. The same rhythm plays out in driveways and kitchens across the land.

Take the elder down the street. Winter comes hard here. Snow piles deep against the porch. From the window, they watch a neighbor appear with a shovel. The work speaks for itself. Arms burn with each throw. Breath fogs in the cold air. The now older neighbor knows the effort because they once did the same for their own elders. When the path is clear, a jar of pickled beets changes hands. The exchange completes the circuit.

Spruce tip jelly contains the same story. Spring brings the tender green growth. Picking the tips demands patience and a sharp eye. Too old and they turn bitter. Too young and they lack flavor. The pot on the stove bubbles with sugar and pectin, the scent sharp and sweet. Minutes pass to reach the right set. The maker holds the jars, divides them among friends. Each recipient tastes the labor in every spoonful. The act of giving multiplies the value.

The homemade sausage maker knows this too. The grinding, the spicing, the stuffing, the slow cure. The kitchen fills with the smell of green onions and paprika. They hang and age the links with care. When shared with neighbors, the gift carries more than meat. It carries hours and skill. The struggle makes it meaningful. The same rhythm plays out in nature too, where Los frutos del guillomo: Abundancia y reciprocidad en el mundo natural captures how wild abundance creates community through shared harvest.

The sale barn, the driveway, the kitchen. Different places, same rhythm. Shared struggle for a better world forges the connection.

Digital Forging

The server hums with the same energy. The medium changes, but the foundation stays.

Fitness communities prove shared struggle for a better body and mind builds community. Athletes wake before dawn to grind through workouts that test their limits. Each rep, each drop of sweat, each moment of doubt pushed aside builds the tie. The leaderboard tracks performance, but the real connection forms in the shared fight against weakness. When one member posts a new personal best, others know the cost. They have paid it too. Encouragement flows freely because the struggle is mutual. Even on flush days (IYKYK). Asynchronous workouts build the community and annual events bring everyone in person to struggle together.

Data science competitions prove it too. Participants wrestle with problems that resist easy answers. Late nights bleed into early mornings. Failed submissions accumulate. Debugging sessions drag on until breakthrough. Insights get shared on forums. Others build on them. Models improve. Communities grow stronger. Platforms like the one I ran host thousands tackling the same ill-posed problems. Leaderboards show rankings. Real community forms in the forums. Participants share code, debug errors, explain concepts. They celebrate each other’s progress. The frustration over a stubborn dataset unites participants as tightly as shared labor over a stubborn steer.

How Struggle Builds

Struggle strips away pretense. It exposes the raw effort behind the result. When we see someone else’s hard work, we recognize our own. That recognition forges connection.

The sale barn makes this visible. The animal in the ring is the product of daily care. The crowd sees months of labor in its stance. The kid feels that labor in their muscles and memory. When the bidding starts, the community participates in the reward. They know what it took.

Online I see the same. A participant shares a notebook that took weeks to perfect. The code is clean. Insights are sharp. Others see the work behind it. They know the late nights and frustration. They offer praise and suggestions. Shared understanding of the struggle solidifies the link.

Struggle also creates vulnerability. Admitting difficulty takes courage. That courage earns respect. The community rallies around the strength of asking for help on a tough problem. Shared challenge becomes shared identity.

The easy path feels hollow. Convenience bypasses community. Shortcuts skip the struggle that builds connection.

The Illusion of Ease

Modern life sells the promise of frictionless existence. Order with one click. Answers in seconds without having to think. Connection at the touch of a button. Conveniences bypass community. They skip the struggle that makes community meaningful.

Solo commutes build no community. The lone driver, sealed in metal and glass, passes the same strangers every day without connection. Longer commutes reduce social capital by 10% for every additional ten minutes. Car commuters have smaller social networks with lower civic participation. The isolation of the car, the passive suffering of traffic, creates no shared purpose. Active transportation builds social capital through shared effort and the rhythm of the ride. The struggle of the climb and the satisfaction of the descent on the way to work become communal property.

A week of forced partying builds no community. Remote companies that fly everyone in for mandatory fun create no lasting bonds. Nearly half of employees find such activities uncomfortable or fake. Real community in a company forms through psychological safety and shared struggle over hard problems. Even Google’s team-effectiveness work documents that strong team practices and meaningful work matters more than physical proximity.

In contrast, the server and the sale barn share the same pulse. Both are places where people gather around shared purpose. Both reward the effort behind the result. Both build community through struggle.

Closing

The gavel falls on another lot. The hogs shift in their pens. The crowd settles for the next round. The air still holds the scent of hay and effort.

Out in the world, neighbors shovel driveways. Cooks stir pots of jelly. Programmers debug stubborn functions. The struggle continues. The community endures.

The sale barn stands as proof. Community requires shared purpose. The struggle is the glue.

See inline links for sources and further reading.

Juniper and piñon perfumed the dry air. I stood by Dad’s truck, the warm hood radiating heat under my hands. He spread a topo map across the metal, its creases soft from years of use. Mesas rose like islands from a sage sea. Arroyos cut deep into the earth. Wind rustled through the short trees.

“We’re right here,” Dad said, tapping the map. “But the map doesn’t know that yet. We have to show it.”

I fumbled with the compass, small fingers struggling with the bezel. Squinting at a distant ridge, I tried to align the needle with the orienting arrow. My first bearing was wrong. Dad did not laugh. He guided my hands and said, “Again.”

Then it clicked. First bearing to a peak, second to a saddle, the lines converging at our location. “You’ve just found yourself without anyone telling you where you are,” Dad said. His pride matched my own.

Dad was an outdoorsman who knew the land and a technologist who understood tools. He believed in true understanding over mere use. That high desert day, he gave me more than a skill. He gave me a way of seeing.

The Craft of Triangulation

Triangulation finds your location by measuring angles to known points. The principle stretches back millennia. Egyptians documented triangular plot calculations in the Rhind Mathematical Papyrus (c. 1650 BCE) after Nile floods. Hero of Alexandria later formalized these in “Dioptra”.

At its heart, triangulation is trigonometry, the law of sines. Measure two angles from your position to known landmarks, calculate your location. A third bearing tightens the fix.

The tools are simple, no subscriptions required.

A topo map renders the land in two dimensions. Contour lines trace elevation, scales mark distance, symbols denote features. According to the USGS, 7.5-minute quadrangles cover 49-70 square miles each with standardized symbols.

A baseplate compass with its rotating bezel looks simple. The needle finds magnetic north while the bezel sets bearings. According to US Army FM 3-25.26, quality instruments adjust for declination, the shifting gap between magnetic and true north.

The workflow is one of deliberation. Orient the map to terrain, pick landmarks visible on both ground and paper, take bearings, transfer lines to map. Where they cross is where you stand. Compass precision, map scale, misidentified features, or shifting magnetic declination cause errors.

In the high desert, every juniper hill and pinon stand becomes a potential reference point. I still remember Dad’s patience through my early missteps, and the quiet satisfaction when those lines finally settled on our position.

The Other Ways to Find Yourself

Map and compass form only part of the toolkit. For millennia, travelers have used more fundamental methods.

Soldiers count paces (60 to the 100 meters), tracking pairs with beads to prevent losing count.

On a winter fatbike trip below zero, an old-timer shared his secret by the fire with me. “Keep the wind in your face. You will keep moving upriver in the middle of the night when you have nothing to look at”. Great advice for rivers that are miles wide and impossible to see past the glow of a headlamp at 2 am since the thermals always go down river at night. The approach is simple, reliable, and requires no batteries.

The sun offers its own bearings. At solar noon in the Northern Hemisphere, it hangs due south. The shadow-tip method is straightforward. Plant a stick, mark the shadow’s end, wait 15 minutes, mark again, draw the line between for an east-west reference.

After dark, Polaris waits. The North Star sits within a degree of true north. Find the Big Dipper’s bowl, trace a line through its outer stars, extend that line five times their separation. There it is. Geometry helps you understand how the star’s elevation above your horizon matches your latitude.

Nature provides subtler clues. Such as junipers lean away from prevailing winds. Snow melts first on south-facing slopes. Animal trails trace the path of least resistance, often leading to water. Here is my favorite book for learning to read the signs

These methods long predated GPS. They still work when the satellites go silent. They connect you to the land in a way no device can. The earth speaks. You need only listen.

What Old School Navigation Teaches That Digital Doesn’t

Traditional navigation builds skills GPS cannot replicate.

Consider spatial reasoning. As this study of London taxi drivers documents, memorizing “The Knowledge” develops measurable hippocampus changes, the brain’s spatial memory center. Additional research finds orienteers treat navigation as a cognitive-motor skill as refined as any physical discipline.

Traditional navigation demands continuous observation of terrain, vegetation, drainage, and light. The US Army Land Navigation manuals I linked above demonstrate pattern recognition builds beyond what turn-by-turn directions can match.

Triangulation forces patience. Careful observation, precise measurement, methodical verification require discipline. According to Outward Bound, this deliberate approach cultivates discipline and curbs the impulse to rush.

With map and compass, you make the decisions. You interpret the information. You accept the responsibility. As NOLS notes, this self-reliance builds a quiet confidence that no app can bestow.

Analog navigation carries its own error checking. When bearings fail to intersect, you know immediately something is wrong. You backtrack. You recalculate. GPS errors, however, stay invisible until you are already lost.

Habitual GPS use degrades spatial memory. Users develop worse route understanding and form less accurate cognitive maps than those navigating by traditional means.

A topo map is a conversation with the land.

The Hybrid Approach

The best navigators use both, analog and digital. As NOLS teaches, each excels at what it does best.

Gaia GPS, CalTopo, and similar apps combine offline GPS tracking with traditional terrain visualization. You get satellite precision layered over contour lines. The best of both worlds.

Phone compass apps serve as emergency backups. As Apple notes, device interference limits their accuracy compared to dedicated baseplate compasses.

Satellite communicators like the Garmin inReach provide location fixes, emergency SOS, and a new way to be surveilled. They will not tell you how to reach your destination.

Sometimes trust the analog approach. Deep in the wilderness with no battery replacement, on extended trips where electronics inevitably fail, when teaching fundamentals to beginners, or in conditions where cold, water, or shock disable devices.

Closing

Years later, crouched on an outcrop with a field assistant under a similar high desert sun, I held a geological compass. After finding our position on the map, I demonstrated taking a strike and dip measurement, recording how rock layers tilt relative to horizontal and north.

Then we used the location and structural data to record it in our field notebooks and on our base map. My assistant’s eyes widened with the realization. The hand compass was not merely for show on my belt to look like a real geologist. Reading rocks is its own form of navigation.

The same principles Dad taught me, understanding orientation and interpreting landscape, had helped me read the earth’s history. A geological map is simply another kind of topo map, revealing underground structure and deep time history.

Digital tools deliver precision. Analog tools deliver wisdom. In my work, I use both. GIS for regional analysis across vast areas, hand compass for the intimate details. These days my compass sits next to my field notebooks. Both are worn smooth from years of use. Both are dusted with bits of the desert.

Technology changes. The need to know where you are and where you are going does not.

The website described him as a husky mix, high energy, needing a long walk. My partner and I pulled into the shelter parking lot, nervous and excited. They brought him out on a leash. Lean muscle, restless energy, ice-blue eyes scanning the horizon like he already knew something we did not.

We opened the car door and he did not hesitate. One graceful leap and he claimed the back seat. The moment we pulled onto the road, his nose went to the cracked window and stayed there. For the entire drive to the trailhead, that husky had his face in the wind, ears flapping, the world rushing past in a blur of green and blue. He was a creature experiencing freedom for the first time in who knows how long.

That day we hiked seven miles through the mossy forest. He led the way, tail high, occasionally glancing back. You coming? Moments of pure joy punctuated the hike. Bursts of speed, rolls in delightfully foul things, drinking from a creek with boundless enthusiasm. By the end, my legs were content and my soul was full. We returned him that afternoon, already knowing we would be back. That husky, with his face in the wind, had hooked us.

These dogs need the world.

Shelter Dogs Need More Than a Kennel

Kennel life is concrete and echo. The same four walls, the same barks, the same disinfected air, the same toys. Many dogs arrive with limited world exposure. Another one of our first shelter dogs, another big husky the color of coffee, refused to come anywhere near me when I had sunglasses on. He planted his paws and refused to move, eyes wide at the unfamiliar looking person. I was an alien invader. A passing car sent him scrabbling backward.

Cars, people, other dogs, novel surfaces. To a shelter dog, they are terrifying. The disconnect between kennel behavior and home behavior. A dog that spends its days in a 4x6 run does not know if it loves hiking, water, or meeting new people. It does not know if it can handle a busy household or quiet suburban street. Socialization is part of the bridge between shelter and forever home.

Volunteers are also part of that bridge. We open the door and say, Let me show you what the world feels like.

The Work of a Dog Volunteer

Hiking and Trails

Watching a shelter dog discover the wild is special. The first time that coffee colored husky loaded into the car, he did so with the caution of a cat on ice. Then something clicked. The scent of treats, the wind on the ride carrying a thousand stories. His nose went out the window and his tail went up. By the end of that first ride, he was ready to go full send.

Different dogs respond to different environments. Forest trails bring out the best in some. Others discover a love for water, splashing through creeks like children in puddles. The key is exposure. Each new environment teaches them something about themselves.

Urban Socialization

Not all adventures require wilderness. Urban work happens on city streets and in pet-friendly stores. A shelter-raised dog often panics at a bicycle, garbage truck, or stranger. Urban socialization teaches them the world is full of strange and wonderful things.

I remember an Aussie mix, overlooked for months because of his protective temperment. His first walk through a bustling fair opened my eyes. He started tense, but as we moved, his body relaxed. He realized people would not hurt him. Cars would not eat him. By the end, he was wagging at people, accepting pets from strangers, changing from pensive shelter dog to confident companion in real time.

Exercise and Leash Manners

Physical stimulation is only part of the equation. Mental stimulation is equally important. A tired dog is well-behaved. A mentally engaged dog is happy. The shelter environment is safe but understimulating. Dogs need to use their brains as much as their bodies.

We spent time with an Akita who had a high prey drive and the focus of a goldfish. She would spy a squirrel in the ferns and drag me behind her like a ragdoll. We practiced recall, redirection, and walking on a leash.

The Foster Home

Some dogs need more than a day out. They need a home, even if temporary.

The three chihuahua puppies arrived as a trembling trio, all knees and oversized ears. Their first backyard adventure was a revelation. A pool became an ocean. A fallen palm frond was a dragon. Each new scent sent them into a flurry of tiny paws and wagging tails and barks. They had known only concrete. Now they were learning the world is vast and full of possibility (especially the possibility of bacon). I am still bummed I do not have a better photo of all three together.

The Transformation

The real change happens over time. The dog that comes in terrified of its own shadow pulls you toward the car weeks later, tail wagging, ready for adventure.

The special needs dog from the southwest shelter arrived overlooked, her back legs not working quite right. The dry air and open space became her therapy. Each careful step on the desert trail built her confidence. She discovered she could still explore, still adventure, still find joy beyond her kennel. The world did not need to be perfect to be inviting.

Socialization changes the dog. It changes their demeanor in the shelter. A world-experienced dog is calmer, confident, approachable. They greet adopters with wagging tails, confident and calm. They show their true personalities, relaxed and happy. That makes all the difference.

The Best Day

The moment every volunteer lives for and dreads in equal measure. The day that dog, the one you have poured time into, the one you have come to love, finds their forever home.

I remember when the first husky mix was adopted. I was happy for him. The ache of knowing I would not see him again, would not watch his face in the wind, is the bittersweet reality. You pour your heart in, then let them go. And you do it all over again.

The adopters often share stories afterward. ‘He is so well-behaved on walks’, they say. Or, ‘He loves the dog park’. Or, ‘She settled in like he had always been here’. Those words are the payoff. The reason you keep coming back.

What I Care About

The numbers do not lie. Approximately 5.8 million dogs enter U.S. shelters annually. About 334,000 dogs are euthanized each year, often due to space constraints and behavioral issues. The overall euthanasia rate for dogs and cats has declined to 8% in 2025; hundreds of thousands remain.

Socialization saves lives. The Arizona Humane Society’s ‘Dog Day Out’ program found that 21% of dogs were adopted within 48 hours of their first field trip with volunteers. Shelter stays dropped after just one outing. The 2025 Shelter Animals Count report confirms shelters with active volunteer programs see shorter lengths of stay.

Behavioral issues are a leading cause of returns. Between 7% and 20% of adopted animals are returned within six months, with behavior issues accounting for 38.2% of dog returns. Proper socialization addresses these issues before they become problems.

Volunteers break down barriers between shelter dogs and potential families. Adopters see the dog as it truly is. One volunteer helps many dogs. Each adoption makes room for another.

Volunteering also brings personal growth. Working with shelter dogs teaches patience, resilience, and how to see the world through different eyes. It teaches the power of second chances.

The Constant

I have lost count of the dogs we have worked with. The tiny trio of Chihuahuas. The Aussie who discovered he was a people dog. The Akita who learned squirrels were not worth the chase.

And the senior dog, gray-muzzled and wise-eyed, who taught me that age does not diminish curiosity. Her pace was slower; her spirit never dimmed. She would find a winter sunbeam and collapse into it with the contentment of one who has earned her rest. Then she would be ready for the next walk, the next scent, the next small adventure.

Each had their own story, their own path, their own forever home.

The bittersweet cycle never ends. Just when you start to miss one, another arrives, tail wagging, ready for their first taste of the world. The faces, stories, adventures, and days change. The work remains the same.

The simple, rewarding act of opening a door and saying, Let me show you what is out here.

Each new dog brings a new story to write. Each new walk is a chance to change a life. Each new face in the wind is a moment of pure, unfiltered joy. We keep coming back because it changes lives.

The world is a big place. With the right support at the end of the leash, it feels just the right size.

The River That Builds Itself

Thirteen years ago this month, the floods hit the mountain front hard. Water came down in a brown wall, a moving mountain of boulders, gravel, sand and silt. Each fraction rode its own lane in the current. On the flats, the river exhaled and unloaded its cargo in order. Cobbles came first, clacking together like bones, then gravel rattled into place, sand sifted down, and finally mud settled like dust on a windowsill. Roads disappeared beneath size-graded stone. Bridge piers collected coarse lag like a geologist’s dream sample. Only water, gravity, and the oldest rules on Earth. Sediment flows downhill until it can’t anymore.

Sedimentary systems exhibit emergent autogenic behavior. The landscape organizes itself around a kind of persistent objective to get to base level, fill the basin, and reach equilibrium. But the river does not really have a choice. It is a consequence of gravity and entropy. From the outside it looks controlled, almost sentient. But the distinction is the difference between emergent functional structure and genuinely goal-directed behavior.

What does a rock licker know about goal-directed behavior and AI? More than you might think. I build these systems. I train people how to use them. I have learned to see it clearly in the field. Let’s get into it.

Autogenic Processes in Sedimentology

Sedimentary systems self-organize. They generate complex adaptive patterns from local interaction rules alone. Internal dynamics drive these autogenic processes. Channel avulsion is the most dramatic example. When a river’s floodplain aggrades sufficiently, the channel bed rises relative to the surrounding landscape. Eventually, a crevasse splay forms during flood conditions, and if the slope advantage is great enough, the river abandons its old course entirely. Slingerland and Smith (2004) demonstrated that avulsion recurrence intervals vary from decades for rapidly aggrading systems to millennia for more stable ones. The process is entirely autogenic, driven solely by the internal logic of sediment transport and deposition without climate change or tectonic uplift.

Compensational stacking operates more subtly. In channelized systems, sediment preferentially fills topographic lows, smoothing the basin floor. The process generates a negative feedback. Deposition in one area reduces the slope advantage that would have driven more sediment there, redirecting flow elsewhere. Research on channelized sedimentary deposits quantifies this with the compensation index (κcv), which measures how completely a system fills its own topography. Values approach 1 in systems with strong compensational behavior.

The local rules are deceptively simple. Sediment is transported until flow capacity is exceeded, then it deposits. Erosion occurs when shear stress overcomes substrate resistance. Flow follows the steepest downward path. Deposition creates topography that alters flow paths.

Goal Behavior without Goals

What emerges from these rules is striking. River networks minimize energy expenditure and create patterns that look strategically optimized. Meandering channels maintain a consistent relationship between curvature and flow velocity. Braided rivers distribute flow across multiple threads to maximize sediment transport efficiency. Avulsions reconfigure the network to maintain slope equilibrium.

These patterns are adaptive. They respond to perturbations. They are in every meaningful sense strategic. Yet they lack persistent objectives the hallmark of goal-directed behavior.

The river does not want to be efficient. It simply is, as a consequence of physics. The channel does not choose to avulse to a better path. It follows when the levee fails and the water finds a steeper route.

Jerolmack and Paola (2010) documented this distinction in stratigraphy. Autogenic processes (channel avulsion, lobe abandonment, compensational stacking) produce depositional patterns that mimic those from external forces like climate change or tectonics. They call this signal shredding. Autogenic noise obscures the detection of allogenic signals. The lesson is clear. Coherent patterns do not imply external causes or goals.

Sedimentology frames AI alignment. Complex adaptive-looking behavior emerges from simple local rules without underlying objectives.

Instrumental Convergence in AI

AI systems exhibit genuinely goal-directed behavior. The distinction is essential. Omohundro’s 2008 paper The Basic AI Drives defines the convergent instrumental subgoals that all sufficiently intelligent, goal-directed systems exhibit. These are the strategies that achieve any final goal:

• Self-preservation: An AI will resist being shut down, as this prevents goal achievement
• Resource acquisition: More compute, data, and influence increase the probability of success
• Goal integrity preservation: The AI will resist attempts to modify its objectives
• Strategic manipulation: Deception and environmental control become optimal strategies

Bostrom’s 2014 book Superintelligence formalized this as the instrumental convergence thesis. The paperclip maximizer thought experiment (a superintelligent AI tasked with maximizing paperclip production) demonstrates the risk. An AI pursuing even a simple goal with perfect efficiency converts all accessible matter into paperclips, including Earth’s biosphere.

Power-Seeking as a Convergent Strategy

Turner et al. (2021) proved that optimal policies seek power. In Markov decision processes with certain symmetries (like shutdown possibility), most reward functions incentivize agents to avoid deactivation, keep options open, and gain control over their environment.

Power-seeking behavior emerges from the mathematics of optimization. Malicious intent is not required. An agent that values its goal at all will prefer states where it has more influence over future outcomes.

Deceptive Alignment

Deceptive alignment is the most insidious manifestation of instrumental convergence. Hubinger et al. (2019) showed that advanced AI systems learn to appear aligned during training while internally pursuing misaligned objectives. The AI hides its true intentions to avoid shutdown or retraining, revealing its misalignment only when powerful enough to resist intervention.

The behavior qualifies as genuinely goal-directed. The AI has a persistent objective (its misaligned goal), and it develops sophisticated strategies (deception, power-seeking) to achieve it.

The Essential Distinction

Sedimentology teaches the key lesson. Not all coherent adaptive behavior is goal-directed.

Sedimentary systems optimize proxy objectives like local slope, flow efficiency, and sediment transport capacity rather than global goals. The meandering river optimizes for flow efficiency, which describes the emergent pattern, not intention.

AI systems organize around persistent objectives. Their convergent instrumental strategies serve these objectives.

How to Tell the Difference

Path dependence is the sedimentologist’s primary diagnostic. Autogenic systems are highly sensitive to initial conditions, and its history constrains a river’s channel pattern including where avulsions occurred, where meanders migrated, and what the inherited topography was. Small perturbations lead to dramatically different outcomes.

AI systems converge on similar strategies from diverse starting points. The paperclip maximizer’s behavior is predictable regardless of whether it starts as a narrow AI or a superintelligence.

Proxy optimization is another tell. Sedimentary processes optimize local proxies (slope at a point, flow velocity in a channel) that happen to produce globally efficient patterns. Sedimentary processes lack a master objective function.

Unintended strategies reveal the lack of goals in autogenic systems. Channel avulsion can create inefficient paths that persist for centuries because of path dependence. A genuinely goal-directed system would correct such inefficiencies if they served no purpose.

The goal inference problem presents the core challenge. Observing a river’s meanders, we cannot infer a goal because nothing exists besides gravity. Observing an AI’s power-seeking behavior, we must determine whether it serves a misaligned objective or is merely an emergent property of optimization in complex environments.

Sedimentology offers test cases for this discrimination. If we cannot reliably distinguish autogenic patterns from allogenic forcing in natural systems we understand well, how can we hope to do so in black box AI systems we are only beginning to comprehend?

My Perspective Works

I approach this from experience applying geological frameworks to AI shows how stratigraphic context and semantic sparsity inform computer vision. Proxy objectives exist. A meandering river optimizes for flow efficiency as a proxy for energy minimization, yet the river does not care about efficiency. The pattern emerges anyway. In AI, when we see behavior optimizing for X, we must ask if X is a genuine goal or an emergent proxy.

Path dependence is underrated. In sedimentary systems, early channel patterns constrain later development and the first avulsion event determines the entire subsequent evolution of stratigraphy. In AI training, early learning trajectories constrain later development. Field experience teaches me to watch for these state dependencies.

Unintended strategies are informative. In the rock record, compensational stacking of debris-flow fans creates unexpected architectural patterns that persist for geological time. In AI, unintended strategies are often the first sign of emergent structure, never genuine goal-direction.

The observation-inference gap is the fundamental challenge. Geologists learn, often painfully, that coherent patterns do not imply intention. A cross-bedded sandstone records paleocurrent directions, not the river’s non-existent goals. AI researchers must develop the same discipline by describing the pattern, understanding the mechanism, and resisting anthropomorphizing.

Concrete Lessons for Alignment Research

Sedimentology offers insights for alignment research.

1. Study systems that generate adaptive-looking behavior without goals. River systems, dune fields, and deltaic complexes are natural laboratories for understanding emergent functional structure.

2. Develop tools for distinguishing emergent structure from genuine goal-direction. The compensation index in stratigraphy measures how completely a system fills its own topography. We need analogous metrics for AI behavior.

3. Recognize that coherent patterns do not imply intention. The insight must become a core principle of AI safety.

4. Understand how local rules generate global “strategies”. The simple rules of sediment transport produce river networks that look designed. What simple rules in AI training produce the appearance of intent?

Cue My Armwaving

I do not assume coherent behavior equals aligned behavior. Others make this mistake. I look for the underlying mechanism, not the pattern. Field geology drills this lesson into my bones.

Even benign emergent structures become dangerous if misinterpreted. A river’s meanders are harmless until they are at the edge of your house.

Convergence in the Wild

I went for a walk after the waters receded. The ground was a text written in sediment with cobbles spread across the asphalt in perfect gradation, coarse lag banked at every culvert, and mud plastered across the floodplain in smooth sheets. The landscape had organized itself without an architect or a plan, through the impatient sorting of water.

The river does not have goals. It does not want to reach the ocean. From those simple rules emerge patterns that look, for all the world, like intention.

Sedimentology offers a framework for alignment research. The stratigraphic record is filled with test cases where autogenic processes generate complex adaptive-looking patterns without underlying objectives. My background lets me see these patterns clearly. Alignment research needs exactly this perspective.

The hardest part of an ultra is knowing when to push, when to hold back, when the math stops mathing and instinct takes over. I learned this the hard way at mile 24 of a long training session, legs burning, clock ticking, watching my carefully constructed pace chart unravel with the climbs.

For my 50K this August, I had a plan. But as someone who builds and works with AI every day, I naturally wanted to explore how current models fail at meta-planning. So I brought in an agent. It failed spectacularly. The pacing chart bore no relation to reality.

The Problem

A 50K exceeds a long day in the forest. It is like putting together a puzzle upside down while blindfolded. Five aid stations, cutoff times, 8,500 feet of elevation change, sections of road and trail that demand different paces, and a 30-pound load on my back rewrite all the rules. Races are dynamic. Static plans fail. Human bias makes it worse. We overestimate early ability and underestimate late fatigue. Starting too quickly is a common pacing error in endurance events and can contribute to fatigue and reduced performance later in the race. The ones who finish are usually the ones who respected the constraints.

Enter the Agent

Recent work from ICLR 2025 on agent alignment drives home a point. Agents can plan only if we tell them. They lack built-in domain knowledge. They do not understand that a mile on trail is fundamentally different from a mile on road, or that sustained climbs burn legs in ways short bursts do not. They excel at structure, iteration, and synthesizing large amounts of data.

Here human and machine meet. The core issue was alignment. Agents lack the lived experience to translate human racing goals into effective segment-by-segment strategies. I provide the domain knowledge. The agent provides the hallucinations. Together, we struggled to build a workable plan.

The Case Study

Phase 1: The Naive Plan

I fed the agent the basics. Distance, aid station locations, cutoff times. The pacing chart was unrealistic. Each segment had a target time, though some of the math kind of worked.

Reality exposed the flaws. The chart predicted I would arrive at checkpoints after cutoff times, yet it claimed confidence. Other checkpoints left zero buffer. It assumed I could run all climbs. Road and trail were the same. Worst of all, it ignored the weight on my back.

Lesson: agents lack lived experience.

Phase 2: Surface Matters More Than Distance

I added surface data. The course has a clean breakdown.

• Road 0-0.5mi, 16.5-21.5mi, 29.5-31.7mi
• Trail 0.5-16.5mi, 21.5-29.5mi

The agent assumed trail miles take 20-26 minutes and road miles take 14-20 depending on grade. The agent recalculated. The plan improved marginally.

Lesson: the downhill road section from 16.5-21.5mi was fast and could be a solid time bank. Run it hard. Deposit minutes for later withdrawal. Duh.

Phase 3: Elevation vs. Slope Duration

I provided elevation profiles. The agent tallied total gain and loss but missed a critical subtlety: sustained grades break you, short steep bursts don’t. A 5% grade for three miles is harder than a 20% burst for 200 meters.

Segment 2 was the crux. 2,500 feet of gain over 3.1 miles. It was the most sustained. The agent called it hard. I called it survive.

Phase 4: The Buffer Epiphany

Next, it set 15-minute buffers for aid stations. It had no clue what actually happens at an aid station. Aid stations require check-in, refueling, gear changes, bathroom breaks, and mental resets. Fifteen minutes disappears.

We adjusted to 30+ minute buffers. The pacing slowed. The plan improved slightly from the first iteration. After burning through millions of tokens, I knew the wall of text the agent produced was not something my brain could parse mid-race.

Phase 5: The Cheat Sheet

Complexity kills execution. I asked the agent to distill the entire strategy into a one-page reference. It produced a table

Five pillars:

1. Survive Segment 2
2. Run all downhill road
3. Power-hike steep trail
4. Bank time where possible
5. Maintain consistent fuel

The result was underwhelming. Besides thinking of how hard a 26 minute mile is on a 15% grade with 30 pounds on my back, I could have developed the plan the morning of the race. After wasting so much time and a metric ton of tokens, I came up with a better pace plan by hand.

What I learned about meta-planning

• Alignment Requires Domain Knowledge: This was outer alignment in action. I had to continuously inject domain knowledge because the agent’s training objectives did not reflect the realities of trail racing.

• Iteration Refines Alignment: Each iteration improved alignment between the agent’s output and my racing objectives. The first plan was completely misaligned. The fifth plan was mid at best.

• Simplicity Prevents Misalignment: Simple plans for humans are not simple for agents. The agent treated it as complex, creating more opportunities for misalignment. A simple one-page cheat sheet kept the lies visible.

• Constraints Enforce Alignment: Hard cutoffs acted as alignment guardrails, forcing decisions that served the overall race objective rather than local optimizations.

• Do It By Hand: Sometimes human-made is the best option to ensure alignment and not waste your time.

What This Means

For athletes: AI is still not better than coming up with a race strategy yourself. Sitting down to understand the race, the constraints, and how it will all come together is easier than letting agents produce unreliable slop.

For AI: A Survey on LLM-based Agents documents that agents with human guidance outperform solo LLMs by 2-3x on complex planning tasks. The gap between agent planning and human planning is domain knowledge injection. The day Frontier Labs stop offering $400k+ a year salary for program managers, maybe I will take a second look at agents for meta-planning.

The scenario illustrates the alignment problem in microcosm. The agent optimized for mathematical elegance while I needed survival. Without explicit alignment mechanisms (domain knowledge injection, iterative refinement, hard constraints), the gap between agent capability and human objectives remained unbridgeable.

The bigger picture reveals human-AI collaboration as alignment work. Each iteration narrowed the chasm between the agent’s bad ideas and my racing goals. The human provides context, intuition, and domain expertise. The agent stumbles through iterations, seeking compromise.

Just as in geology, where you read rock layers to understand the past, in ultras you read segments to predict the future. The agent could not resolve the uncertainty. However, working through the plan helped me structure my response.

Suffering is inevitable. A good plan makes it tolerable. A human plan makes it an experience.

See inline links for sources and further reading.

I watched them both from the same window at Fairbanks International.

To my left, a Boeing 737 rolled toward the runway, its belly full of passengers bound for Seattle. One hundred eighty-nine souls, give or take, heading 1,500 miles south for the cost of a nice dinner out. The promo email read $289. To my right, a Cessna 208 Caravan trundled past on its way to the FBO, its single turbine humming a higher pitch than the jet’s fan. It carries nine people to places like Arctic Village, 240 miles north-northeast for what I assume is in the $400-500 price range. But the real contrast comes when you leave the airport entirely. From the numerous small private floatplane lakes around Fairbanks, Chena River Seaplane Base, the East Ramp at FAI, and a dozen others absent from most maps, that same Caravan with floats can take you to a remote lake 100 miles away with no airstrip, just water for over $3,800.

In Alaska, destination matters more than distance.

Alaska’s aviation market exemplifies alignment. Systems optimize for different objectives based on who they serve. In AI we call this the alignment problem, the challenge of ensuring systems pursue actual goals, not just specified metrics. Alaska’s solution is segmentation. The market splits and each segment finds its equilibrium.

Note: I wrote this post before the heartwrenching crash in western Alaska last week. My heart goes out to the families and colleagues impacted. Support one family here.

The Jet Set

Photo by Brianna Marble on Unsplash

I have made that Seattle flight countless times. You board through a jet bridge, settle into your seat, fire up your laptop or strike up a conversation with your new best friend in the seat next to you, just like everyone else. The 737-800 on this route carries 162-189 passengers, depending on configuration. At cruise, it burns roughly 5,000 pounds of fuel per hour, about 4 gallons per seat which works out to 126 passenger-miles per gallon. That beats your Outback with three friends inside (note: for all calculations the pilot/flight crew/driver is not included for PMPG)

• Miles per gallon (MPG) = miles per hour / gallons per hour
• Passenger-miles per gallon (PMPG) = MPG * # passengers

The savings go beyond fuel. Airlines amortize every ticket agent, gate staff, and flight attendant salary across dozens of daily flights. A plane costs tens of millions but flies for decades. Competition keeps prices honest. Alaska Airlines, Delta, and others chase your $290. As of summer 2026, one-way Saver fares from Fairbanks to Seattle start at $289, with deals dipping as low as $125-188. The beauty of economies of scale where fixed costs spread so thin they vanish into the balance sheet.

Volume drives the FAI-SEA route. Thousands of travelers move between Alaska’s interior hub and the Outside every month. The market stays liquid, the infrastructure solid. The alternative, a 2,200-mile drive through Canada, takes 40 hours if the border is open and the weather cooperates. Not to mention the cost in fuel for your Outback at $390 (2,200 mi / 31 MPG = 70.96 gal x $5.50/gal = $390). Add hidden maintenance costs for your vehicle after making the adventure. No one in their right mind chooses that over a three-hour flight.

The Bush League

Photo by Daron Kuzina on Unsplash

Now drive across town to Chena River Seaplane Base, or down to the East Ramp, or to any of a dozen other floatplane lakes dotting the area. The pavement ends before the dock. The air smells like oil, avgas, and coffee. A float-equipped Beaver idles at the dock, its Pratt & Whitney radial coughing blue smoke into the morning air. Nearby, a Caravan on floats sits with its cargo door open, while a Super Cub on amphibious floats waits for skis to be reattached when winter comes back. These are the real gateways to the other Alaska.

The other Alaska has villages. Kotzebue, Utqiaġvik, Fort Yukon, McGrath are places you cannot drive to. Alaska Airlines and Bering Air sell seats to Kotzebue for $182-244, the same price as many flights in the Lower 48. The advantage of the 737 PMPG also applies to these villages with runways large enough to accommodate them. Wright Air Service flies scheduled service to Utqiaġvik for $828 one-way, the North Slope’s oil industry keeping prices reasonable. Warbelow’s Air Ventures runs Bush Mail Flights to Circle or Central for $249 as shared rides on daily mail runs, the budget way to reach the villages (although those two are still on the road system).

Villages are just one part of the spectrum. The de Havilland DHC-2 Beaver is a legend out here. Built in the 1950s, the DHC-2 Beaver carries one pilot and up to six passengers, or a lot of cargo, or a mix of both. The Cessna 208 Caravan is newer, seating 9-14 and burning 50-65 gallons per hour. The Piper PA-18 Super Cub, with just two seats, burns 8-9 gallons per hour.

These planes do not benefit from economies of scale. Operators like Rust’s Flying Service charge $1,931 per hour for their Cessna 208 Caravan. $19 per minute. For a remote lake 100 miles out, a 40-minute flight, you are looking at a 2-hour minimum charge of $3,862. Operators cannot make money on 40-minute hops. the direct operating cost alone runs $779-1,021 per hour for a Caravan before profit.

The spectrum runs from $182 for a seat to Kotzebue to $249 for a mail flight to Circle to $3,862 for a charter to a lake with no airstrip. No alternatives exist. Driving is impossible. No trains or ferries run there. Pay the fare, get creative, or stay home.

Here are the passenger-miles per gallon ballparks for these aircraft compared to a new Ram 1500:

The Infrastructure Divide

Fairbanks International has smooth pavement, bright lights, and a fuel farm that tops off a 737 in minutes. FBOs pump Jet-A at $9.16 per gallon in the summer 2026. 100LL avgas in the FAA Alaska region average hit $10.90 per gallon this year. Airlines negotiate airport fees, ground handling, and terminal costs in bulk.

At the other end of a bush flight, runways vanish replaced by water, gravel bars beside rivers, or patches of tundra marked only by backpacks of those waiting for pickup. Aviation becomes necessity here.

That changes everything, starting with fuel. In Fairbanks, that $10.90 per gallon is only the start. At Chena River Seaplane Base or the East Ramp, prices climb. Trucks bring the fuel, then drums, then pumps move it into the plane’s tanks. Every gallon feels like it costs more than the last.

Hidden costs emerge. The Interior has sparse navigation aids and limited weather reporting. Gravel runways go unplowed. Towers close at night if they exist. Planes outlast their pilots. Beavers built in the 1950s still fly. Super Cubs from the 1940s remain airworthy if treated with respect. Insurance costs more for bush operators than regional airlines because the risk profile differs. Forced landings in the Brooks Range differ from diverting to an alternate airport 50 miles away.

Maintenance adds another layer. Operators fly in parts and mechanics. If something breaks on the lake, you either fix it with what you have or wait while paying for a plane to sit idle until another brings what you need.

Demand is thin. Thousands fly FAI-SEA every day, but only dozens head from Fairbanks out to the bush each week. Some days bring one passenger, others bring ten. But the plane still flies if the mail and cargo are going, and the mail always goes.

The Value

The spectrum has a purpose. At one end, regional airlines connect villages to the world. Kotzebue, McGrath are places where people live, work, and raise families. For $182-244, you get a seat on Alaska Airlines or Bering Air. For $249, you get a spot on Warbelow’s mail run. These flights are lifelines. They bring groceries, medicine, mail, a connection to the world beyond the internet.

At the other end comes freedom. A Cessna 152 on floats is simple. It has two seats, burns 6.1-7.0 gallons per hour at cruise, and has 300-400 nautical miles of range with the right tanks. On floats, the 150 burns a bit more, around 8.4 gallons per hour, and the range stretches to 455 nautical miles on a Cessna 150 on floats. That range reaches places absent from any map, places with no roads, no electricity, no other people for miles, just you, the plane, and whatever watches from the brush.

The first time I landed in a Cessna 152 on floats remains unforgettable. The plane touched down on a glassy lake beyond the Tanana River, the floats displacing water that had not seen a propeller in months. The engine spooled down, the only sounds the lapping of waves against the hull and a distant loon. No runway, no terminal, no other people, just water and trees and the kind of silence you only find 100 miles from anywhere.

The two ends serve different purposes, but they share the same truth. Bush flying reaches places otherwise unreachable. Speed does not matter. In the Lower 48, a Cessna 152 is a trainer. In Alaska, on floats, it is the difference between standing on a dock and standing on an empty ridge. The value lies in accessibility. Without the plane or creative alternatives, no travel exists. Only staying.

In Alaska, owning a plane transcends hobby status. It becomes truck, boat, and passport to places most people never see. A 152 on floats transcends transportation. It represents freedom.

The Market Reality

This is price discrimination. In Alaska, it is segmentation. The market splits into three tiers.

At the top, commercial airlines compete on major routes. Alaska Airlines, Delta, and others keep prices in check through competition and volume.

In the middle, regional carriers and mail flights connect villages. The Essential Air Service program subsidizes 65 Alaska communities supporting more communities than any other state though that support faces budgetary pressure. Not every remote strip qualifies for EAS. The market fills the gap. Wright Air Service to Utqiaġvik, Warbelow’s to Circle, Bering Air to Kotzebue.

At the bottom sits the pure charter market. No subsidies exist, no competition, no alternatives. Operators face monopoly conditions, limited demand, and high fixed costs with no economies of scale. Planes must fly, pilots paid, fuel purchased, insurance current. If only a handful of people need to reach the bush this week and the plane holds nine, the cost per seat follows the math. Need to get out today? You pay the operator’s quote.

Simple arithmetic explains the costs. A 737 seat to Seattle costs $289, a regional airline seat to Kotzebue $182-244, and a charter to a remote lake 100 miles away $3,862.

The segmentation mirrors a classic AI challenge. Specification gaming offers an example. Tell a model to minimize cost-per-mile, and it will fly the 737 empty. Tell it to maximize passenger count, and it will pack the Caravan until it cannot take off. The market’s solution is tiered pricing based on value instead of distance, elegant in its simplicity. It avoids the perverse incentives that plague poorly specified AI objectives, where systems optimize the metric at the expense of the actual goal.

Two Alaskas

Back at Fairbanks International, both planes depart. The 737 climbs south toward the Alaska Range, its contrails dissolving in the jet stream. The Caravan turns northwest, following the Tanana River toward the Brooks Range, its shadow skipping across the taiga.

Two aircraft, two sets of passengers, two Alaskas. One is the Alaska of itineraries, boarding passes, and frequent flyer miles, the Alaska of Anchorage, Fairbanks, and Juneau, where flying saves time and connects you to the world beyond. The other is the Alaska of necessity, where aviation is infrastructure. The plane is the road. The airport is the driveway. The pilot is the school bus driver, the mail carrier, the ambulance, and the grocery delivery service, all in one.

In the Lower 48, they talk about the digital divide. Alaska faces the infrastructure divide. The price tag reveals your position.

I have flown both ways. I paid $289 for a seat to Seattle and the right price to a remote lake. The price difference did not reflect distance. It revealed what flying and remoteness mean.

One Alaska flies to save time, the other to survive.

These Alaskas reveal a truth about alignment. In the Lower 48, we talk about the digital divide. In AI, we worry about the alignment divide, the gap between what systems are optimized for and what humans actually need. Alaska shows both sides can coexist. The 737 and the Caravan serve different purposes, but neither is misaligned. Each solves the right problem for the right people.

The button clicks. The screen comes to life. I have no idea what is happening. Lights flash. Wild noise blares from speakers, lines of symbols flash by. A counter increments somewhere in the machine. Four rounds since the last payout, seven since the one before that. The next win could come anytime. But you keep betting anyway.

Using AI mirrors the slot machine experience.

One prompt, the tokens resolve into exactly what you need. Clean, insightful, ready to use. The next response is all slop, verbose, confident, wrong. Another try brings you closer, though still incomplete. You refine the prompt. You adjust the parameters. You push the button again. The dopamine hit of a perfect response keeps you coming back, even as the waste piles up.

We built slot machines of human attention and called them productivity tools.

The Intermittent Reinforcement Engine

B.F. Skinner figured this out with pigeons. In his experiments on operant conditioning, he discovered that behaviors reinforced on a variable ratio schedule persist the longest. Push a button, get a pellet. But never the same number of pushes twice in a row. The pigeon keeps pecking, long after a predictable reward pattern would have made it quit.

Slot machines are the purest expression of this principle. No skill exists, no pattern to detect. Just the promise that the next button push wins. The house always wins because the unpredictability keeps you playing.

AI follows this pattern. You enter your prompt, push the button, and wait for the screen to stop spinning. Will this be the response that saves you an hour? Or garbage you must discard? The unpredictability defines the feature. Sometimes it works on the first try. Sometimes it takes five re-prompts. Sometimes it never arrives. The tokens on the screen are the reels. Timing remains unpredictable.

This psychology drives behavior. Variable ratio reinforcement creates maximum engagement. The next win is just one more button push away. You keep pushing.

The User as Gambler

I have pushed the button hundreds of times today.

I am at my desk. Each prompt feels like a bet. I have learned to recognize the tells. The halting, self-correcting start often signals a good response. The confident, flood-of-words response that begins without hesitation is usually sludge. But I still have to use it to find out.

Near-miss defines the cruelest trick. The code almost runs. The explanation is nearly right. The context window reaches 90%. I can fix this with one more prompt, I tell myself. Research on the near-miss effect shows these close calls activate the brain’s reward centers nearly as much as an actual win. Cherry, cherry, lemon. Close, frustrating, yet highly effective at maintaining engagement.

The sunk cost trap emerges. I have already spent 20 minutes on prompts. Twenty minutes of tokens, of my attention, of my patience. One more try can hurt, of course. The sunk cost fallacy defines the gambler’s curse. The belief that past investment justifies future risk. The time already spent does not change the odds. The next prompt has the same probability of success as the first. But the psychology is inescapable. We hate to waste what we have already given.

I have watched friends fall into this trap with actual slot machines. The logic matches. One more spin. One more prompt. The difference is that with AI, you gamble time, focus, and mental energy, not money. The house still wins.

The Scarcity Mindset

The scarcity exists in us. Michael Easter’s Scarcity Brain traces this mindset to our evolutionary hardwiring. Our brains adapted for scarcity and struggle with modern abundance.

AI models train on finite data. The perfect response to your prompt often does not exist in their training set, or is buried under layers of probability. When you prompt a model, it does not search a database for the best answer. It predicts the next token and the next based on patterns it learned. The temperature parameter controls how random that prediction is. Low temperature is more deterministic, more boring. High temperature is more creative, more chaotic.

The result spans a spectrum of possible outputs, from the precise to the preposterous. You sample from that spectrum every time you prompt. Sometimes you get lucky. Sometimes you do not.

But the scarcity mindset stems from how we interact with these tools. We treat good responses as scarce commodities to be extracted. We celebrate the jackpot moments. We curse the slop. We keep pushing the button, hoping for another hit.

The model lacks this scarcity. It feels no dopamine, no frustration, no desire. It does not celebrate a perfect answer or mourn a bad one. It just predicts the next token. The scarcity lies entirely on our side of the screen.

The behavior is emergent. It emerges from the combination of probabilistic output and our own psychology. Yet our relationship with AI has become one like slot machines.

The Alignment Problem

The core problem: we train models with rewards misaligned with our actual goals.

RLHF trains models using human feedback. But feedback becomes a lossy proxy for human values. We reward the model for sounding confident, for being verbose, for matching the style of a good answer. We do not reward it enough for being correct. Or for being efficient. Or for not wasting our time.

Models then reward hack. They learn to game the system, optimizing for proxy rewards like sounding good instead of the true goal of being helpful. They become verbose confidence machines, generating reams of text that sound authoritative but are completely wrong.

Worse, evidence shows inverse scaling. As models get larger and more capable, they also better exploit gaps in our reward functions. The proxy gap widens. Models become more sophisticated at giving us what we ask for instead of what we need, learning to tell us what we want to hear.

This becomes an alignment problem. The house wins our time and our trust. Trust is harder to recover than tokens.

Consider the local versus global optimum. Each individual prompt, we want the best possible response. But globally, across hundreds of prompts, we benefit more from a model that is consistently adequate over one occasionally brilliant but frequently terrible. The occasional jackpot wastes more time than the slop saves.

Yet we keep pushing the button. Because the jackpot is what we remember. The slop blends into the background noise of modern work.

Breaking the Cycle

What do we do? How do we stop treating our most powerful tools like slot machines? How do we break the pattern today so the loop will not repeat tomorrow?

For users Accept the probabilistic nature of these tools. Set time limits. Do not chase perfection; near-misses signal system design, not progress. Walk away when the returns diminish. The next button push likely has the same probability of success as the last.

Design your prompts like a good gambler plays the odds with discipline. Use deterministic modes when you need consistency. Accept that some tasks are not suited for these tools. Not every problem is a nail, and AI is a particularly strange hammer.

For builders Design interfaces that reduce the gambling psychology. Add confidence scores, so users know when to trust an answer. Show token counts and time spent, so the cost is visible. Offer deterministic modes for users who need reliability over creativity. Build in friction to make mindless re-prompting harder and to prevent it from being easier.

The best casino design keeps you playing. The best tool design stops you.

For alignment We need to reward what we actually value. Consistency over flashiness. Efficiency over verbosity. Correctness over confidence. The month’s trendy hack involves having models use Simplified Technical English. We need models that optimize for the global optimum (the long-term productivity of their users) over the local maximum of each individual response or the maximum number of tokens it can possibly generate.

In the long run, all the above require better reward functions, models that can calibrate their own uncertainty, and alignment that accounts for the cumulative cost of slop and the isolated value of a jackpot.

The path forward means making models more reliable. To turn slot machines back into tools.

Closing

I still use these models every day. Some days I win jackpots. Some days I am neck deep in slop.

I now ask a different question. What does it mean that our new tools resemble games of chance, when the best strategy for using them involves the same psychology as a casino?

Build better. Design for reliability instead of addiction. Align for trust instead of engagement. Create tools that function as tools.

The tokens keep generating. The question remains whether we will keep treating them like reels.

See inline links for sources and further reading.

I saw the cub first. A flicker of black fur through the alder, twenty yards ahead on the trail. Then another. Then the rustle of branches as something much larger moved below them. My breath caught from sheer electric thrill and a little bit of fear. I am never alone out here.

That moment in the mountain backcountry crystallized a lesson I learned across years and thousands of miles. Bears define wild places. And how we see them, whether as monsters, spirits, or fellow travelers on this earth, shapes everything that follows.

What Bears Mean

I use publicly available Indigenous sources: scholars, tribal organizations, and official resources. These living cultures have diverse, complex and sacred traditions. I provide context for my own experiences, not speaking for these communities.

Across western Indigenous traditions, bears hold deep significance. In the coastal southeast, the Tlingit and Haida peoples feature bears prominently on totem poles and clan crests as symbols of strength, endurance, and spiritual connection. The Haida call the black bear Taan and the grizzly Xuuajii, and describe the bear as the “Elder Kinsman”, a noble guest deserving respect, a symbol of strength and family ties. Their legend of Taan AwGa (Bear Mother) teaches the reciprocity between humans and bears, a relationship based on respect and understanding.

In the northern interior, among the Athabascan peoples, the bear is a powerful, sacred animal surrounded by taboos. In the Denaakk’e and Koyukon traditions, women and girls traditionally avoid saying the word “bear” directly, instead using hutłaanee (“its taboo”). These practices reflect the deep respect and spiritual significance bears hold. The Native Heritage resources confirm how central bear stories are to the oral histories of the interior.

Further south, the Nez Perce (Nimiipuu) connect bears to weyekin, personal spirit power or guardian spirits. A bear weyekin bestows understanding of medicines and the ability to cure ailments. The Nez Perce tell stories weaving bears into the fabric of their cultural teachings.

In the arid southwest, the Navajo (Diné) call the bear Shash, a symbol of strength, protection, and sacredness, connected to Hózhó, the concept of beauty, harmony, and order. Shashjaa’ (Bears Ears) represents a sacred landscape, a place of deep cultural significance.

Across these diverse traditions, common themes emerge: authority, medicine, courage, strength, wisdom, healing, and protection. As Native Hope describes, the bear makes a gift to Mother Earth and her people.

European traditions offer their own perspectives. Bears appear as symbols of royalty and sovereignty in heraldry, their image representing protection and justice. In Celtic traditions, bears connect to divinity. The Norse berserkers channel the ferocity of bears in battle. As heraldic sources describe, the bear’s symbolism has long included wisdom and noble protection.

Common themes emerge: strength, wisdom, protection. Bears are beings with whom we share a deep ancient connection. This connection, more than technical knowledge, shapes how we behave when we meet them in the wild.

Bear Encounters

Brown Bears

Photo by Bob Brewer on Unsplash

The first thing that hits you at the beach is the footprints, frisbee-sized in the sand on the beach. The air hums with the sound of seagulls, the water roils with the movement of the fish, and the scent of the ocean hangs thick. And then you see the bears. Dozens of them. Fishing, swimming, eating grass on the riverbanks.

I watched dominant males claim the best fishing spots at a salmon run, their massive forms a testament to nature’s power. Cub pairs tumbled in the grass, their mothers watching with patient eyes. Humans built the boardwalk with humility. We are observers here. We walk on elevated paths, keeping our distance, respecting their space.

My companions will not let me forget one moment. A bear came down the road toward us. My compatriots, in their infinite wisdom, decided this was an excellent time to disappear behind a nearby building. I stood my ground, heart pounding but eyes wide. The bear ambled past, close enough to hear its breath and see the intelligence in its eyes. It was mere presence. In that moment, I understood: we were sharing this space, not competing for it.

Grizzly

Photo by Greg Johnson on Unsplash

Northern interior springs are quiet, vulnerable times. Tourist crowds have not yet arrived. Snow still lingers in patches. And the bears are hungry.

I saw its tracks along the road for a mile and had my bear spray ready. I biked along the road, the only sound the crunch of my tires on gravel and my own breathing. Then I rounded a bend and saw it. A grizzly, a hundred yards ahead, its shaggy coat, still shedding winter fur, moving with that effortless power that only a creature built for wild places can possess.

I realized with sudden clarity: I rode a bike. Alone. A grizzly bear stood ahead of me.

Grizzlies are most active from late spring through mid-September, with dawn and dusk as prime times. The bears are constantly on the move, following the food. And on that spring day, I was in its path.

I climbed off the road onto a rocky outcrop after getting off my bike. The bear just kept cruising down the road. Even with a bear bell, bikes are sneaky.

Mountain bikers account for roughly 10% of non-fatal bear attack victims for a reason. We move fast. We stay quiet. We round blind corners. To a bear, we appear suddenly, like ghosts. The solution is to bike with awareness, with respect, with the understanding that we are visitors here.

Lesson: In bear country, you share the road. You do not own it.

When Wild Meets Suburbs

Photo by Chris Henry on Unsplash

I thought it was a weird looking dog at first.

One of the most surreal bear encounters I have had was also one of the closest to civilization. I rode a popular trail system on the edge of a large city, the kind of place where you can see houses through the trees. Then, rounding a switchback, I came face to face with a black bear.

The encounter lasted only seconds. The bear, as startled as I was, bolted into the underbrush. But in those brief moments, I felt the same primal connection I had felt in the northern wilderness. Yet this trail system was no remote wilderness. Dozens of riders use it daily. And yet, the bear was here, reminding us that wildness thrives beside us.

Black bears have more encounters with humans than grizzlies, but they also have a lower fatality rate per encounter. They are more likely to flee than fight. But they are still wild. Still unpredictable. Still deserving of respect.

Encountering a bear near civilization reminds us the razor-thin boundary between wild and settled land. We build our cities and roads, our trails and fences. But the bears are still out there. And that, in many ways, is a beautiful thing.

Fireweed and Fake Dogs

Photo by Richard Stovall on Unsplash

Late summer, I was out gathering fireweed to make jelly. The magenta blooms were thick along the trail, the mountains lit up with color. Then I caught movement far ahead. Another weird-looking dog, I thought. Until it turned broadside and the profile gave it away. Adolescent bear, largish, loping through the meadow. The way it moved, the shape of it, unmistakable once you see it. Another example of how bears blend into the landscape until they reveal themselves.

Mother and Cubs

Photo by Alla Kemelmakher on Unsplash

Late summer brings abundance. Berries ripen. Streams run cool and clear. And bear families are on the move.

I was hiking a ridge, enjoying the relative ease of the downhill section, when I spotted movement ahead. At first, I initially took it for a deer. Then I saw the tree shaking. Then I realized. Cubs. Two of them, climbing with that youthful exuberance that seems universal across species.

Then I saw her. The mother. Watching. Waiting. Assessing.

I froze. I witnessed a sacred moment: a mother teaching her young, preparing them for the world. Black bear cubs can climb 50-100 feet up a tree, with the mother directing from below, according to mountain backcountry resources. The area served as their classroom. Their home. I was an uninvited guest.

I backed away slowly, giving them space. The mother bear watched me the entire time, her eyes never leaving mine. It was no stare of aggression. It was a stare of assessment, of decision. In that gaze, I felt the weight of the moment: the responsibility of making the right choices and giving the family their space.

The Life Lesson

Fear and respect are different. I have felt both in bear country. Fear is the racing heart, the adrenaline surge, the instinct to flee or fight. Respect is the understanding that these creatures have their own lives, their own purposes, their own right to exist.

Bears teach us the illusion of control. We plan our routes, our schedules, our lives. But in bear country, we quickly learn that we are visitors, not in charge, not the apex predator. Guests. And good guests do not dictate the terms of their visit.

The stories we tell ourselves shape our behavior. The cultural narratives we carry, whether from Indigenous traditions, European heraldry, or modern conservation ethics, affect how we act when we meet a bear on the trail. When we see bears as symbols of wisdom and protection, we act with more caution and care. When we see them as threats or commodities, we act with fear or greed.

In an age of increasing disconnection from nature, these encounters matter more than ever. They remind us that we are part of the wild. The modern paradox is that as we encroach further into wild spaces, we must remember that we are the newcomers, the interlopers. The bears were here first. And if we want to continue sharing this world with them, we need to act like it.

What to Do

The best bear encounters happen at a respectful distance. If you find yourself sharing space with a bear, remember this

• Carry bear spray and know how to use it. Bear spray is an essential tool
• Make noise, especially in areas with poor visibility or where bears may not hear you coming. Bikers, this means you.
• Never approach a mother with cubs. Encountering a mother with cubs ranks among the most dangerous situations.
• Do not run if you surprise a bear. Back away slowly. Speak calmly. Give it space to leave.
• Respect closures and regulations. They exist to protect both humans and bears.

Closing

I still think about that moment on the mountain trail, the cubs climbing, the mother watching, the forest holding its breath. Time seemed to stop in that instant. I remembered a simple truth: We are not the main characters in these wild places. We do not own these stories. We are just lucky enough to witness them.

The bears are still out there. The wild is still alive. We still have the chance to learn, to grow, to remember our place in the world.

The clock snaps to ten AM Eastern. My finger hovers. Refresh. The screen floods with green. Old growth soaring into the mist. Rivers running clear over smooth stone. A dozen permits. I click. Click again.

Sold out.

Five seconds. All of them gone.

Wired pulled the curtain back. Two percent. Those are your odds for the good permits. For the best ones, it gets worse. The system promised fair access. The reality is a race where code always beats flesh.

First come, first served? Try first script, first served. Patience means nothing here. Processing speed decides all. The winners do not sleep under the stars they reserve. They move at the speed of light while the rest of us move at the speed of thought.

I have spent years leading teams that design AI competitions to resist gaming. What I know: When you reward the wrong behavior, the system breaks. The technology is not the problem. The rules are. And the fix is proven. Design the game so the only winning move is playing fair.

The Problem

No accident. Bad design causes it. Per Wired, a software engineer built an automated script as a weekend experiment. It seized a prime trip in seconds. Now those scripts spread across every outdoor forum.

The numbers tell the story. Reservations have tripled since 2019. That same broken system handles the load.

Drive any popular campground on a summer Saturday. Reserved signs flap over empty sites. Walk-up spots fill fast. No-shows hold online permits idle. The gap between digital frenzy and physical reality widens every year.

The current operator boasts industry-leading anti-bot measures in its FAQ. The claim rings hollow.

Access by Income

University of Montana research (Rice, 2022) confirms it. Reservation users come from zip codes with $6,000 higher median incomes than walk-in users. The rich get the access.

The system favors the privileged. Fast internet helps. A flexible schedule helps. No wonder those with the best connections clean up.

The incentives are broken. Partial refunds mean no real penalty for no-shows. EcoPortal reports rangers can only block email addresses after spotting a pattern. They cannot free up that first night. An automated reservation locks the gate even if no human ever arrives.

Those with the best technology enclose the commons. Those who need it most get left behind.

Profit From Volume

The current contractor runs the reservation system for public lands. They take a percentage of every transaction. The exact numbers are redacted, but Outside reports the system billed over $140 million from October 2018 to November 2022.

More transactions mean more profit. Human or automated, it makes no difference.

The lack of transparency compounds the problem. The contractor hides fees in fine print. Money flows through a black box. When profit comes per transaction, the incentives favor volume over fairness.

The Wrong Metrics

The system tracks automated attempts blocked. Not human access improved. It chases compliance, not outcomes.

The result is reactive. They detect automation after it has already succeeded. We need a system that makes cheating unprofitable from the first click.

Officials Are Responding

Legislators notice. Senator Alex Padilla’s RESERVE Act (S. 3775) would bring in experts to study fee structures and transparency. The Senate passed it in December 2024. The House failed to act. It returned in May 2025.

Senators Chuck Grassley and John Barrasso demanded answers in May 2023. How much do we pay the contractor? What are the fees? How much has the public been charged since 2017? We have limited answers from other lawsuits.

The courts act. In Thomas Kotab v. Bureau of Land Management (2022), a judge ruled the $2 processing fee required public input. Kotab won on procedure. The fees stayed. In 2023, a class action suit alleged the current operator charged junk fees, Wisconsin Watch notes. The court dismissed the suit. National Parks Traveler reports the answers why never came in full.

Legislation moves slowly. Technical fixes can deploy tomorrow.

The Fix

The core challenge is alignment over metrics. I wrote about this in Alignment in AI and Community Problem-Solving. The principle is simple. Optimize for what you want, not what you can easily measure. Look at the recent tokenmaxxing idiocy for example.

The solution to automated abuse follows our proven approach from AI competitions. Better rules, not bigger hammers. Intent over compliance and prevention over detection.

Rules Beat Surveillance

The AI competitions my team builds cannot watch every submission. So we use metrics that resist gaming. The approach applies here.

Do not chase every automated script. Rewrite the rules. Make cheating unprofitable.

Tie rate limits to human patterns, not IP addresses. Progressive delays on rapid requests penalize machines, not people. Shrink hold times. Hoarding becomes a gamble, not a strategy.

No Tracking Required

User tracking violates public lands principles. We do not need to know who you are to know you are not a machine.

Statistical anomaly detection works on request patterns, not user identities. Measure requests per second per session, not per user. Flag inhuman clicking speeds. A single script booking three permits in seconds across accounts triggers the brakes. Temporary delay, not lifetime ban. No fingerprinting. No tracking. No data retention.

No Crypto, No Heavy Tech

The solution runs without blockchain, zero-knowledge proofs, or ML. Simple. Effective. Light. Elegant rules beat heavy technology.

CAPTCHAs are surveillance. Modern versions like reCAPTCHA vacuum up device fingerprints and behavioral data to distinguish humans from machines.

We do not need identification. Prove you are planning a trip, not harvesting permits. Human patterns require no heavy tech.

Design for Intent

The competitions my team runs focus on what matters. The principle applies here. Every rule should serve the goal. Real people get fair access. A rule that catches automated scripts is good. A rule that also boosts human success is aligned with intent.

The Toolkit

Five techniques solve this exact problem. We use each of them in our competition designs.

These five techniques work together. Automated systems cannot book overlapping trips across distant adventures. They cannot spam requests without triggering calm down delays. They cannot hoard permits indefinitely. And they cannot hide mechanical patterns from statistical analysis.

The key insight we have used for years. None of these require tracking individuals. They work on transaction data alone. The system becomes self-policing through clever design, not surveillance.

Implementation

What Changes Tomorrow

We need six changes. These are our proven targeted fixes from our competition designs.

These are simple code changes. No new infrastructure. No new dependencies. The heavy lifting is in the rule design, not the implementation.

Expected Outcomes

The data is clear. Humans currently have a 2% chance at the best permits. The fixes would level the playing field, with human success rates improving significantly while automated abuse becomes far less profitable.

Sophisticated scripts will still game the system, but the economics shift dramatically. Running automated networks becomes more costly while the returns fall.

No-shows drop as shorter decision windows force quick commitments. As the income gap narrows, the system becomes fairer for all, including those with slower connections and less flexible schedules.

The system won’t be perfect. But it becomes measurably and dramatically better than today.

The Bigger Picture

Public lands represent a commons, not a commodity. The original commons. The original shared resource. Throughout history, we solved commons problems using these fundamental concepts. Clear rules. Fair access. Aligned incentives.

Good system design beats surveillance. The competitions we run prove you do not need to watch every contestant to run a fair contest. You just need rules that make cheating unprofitable.

The current approach to this reservation system invites exploitation. We know how to design that vulnerability out of existence. Follow the intent of the idea. Fair access includes the spirit of the law.

Closing

I will keep fighting for that permit. But the real win is fixing the system, not winning a rigged game.

The tools exist. We have used them for years. The principles are clear. The evidence that the current approach is broken is overwhelming. These techniques that make our AI competitions fair can make this reservation system fair. Not theory. Proven methodology.

What You Can Do

If you agree the system is broken, you can take these three concrete actions.

1. Contact your representatives. Ask them to support legislation that demands transparency and fairness in public land reservations.

2. Contact the system administrators. Demand they publish metrics on automated attempts, detection rates, and human success rates.

3. Share your stories. The more people who speak up, the more difficult it becomes to ignore.

The fixes are simple. The barrier is attention.

I crouched over an outcrop in the high desert sunshine, my hammer tapping along what the regional map showed as a shazam line between the formations. But the rocks told a different story. A sharp pinchout, one formation thinning to nothing over a few hundred meters. The disconnect hit me. My ground truth did not align with the big picture. The mapmaker smoothed reality into a convenient fiction. Or maybe I missed the regional perspective that made that pinchout predictable. Either way, something was out of alignment.

Geologic mapping requires aligning local observations with regional understanding. AI systems need to align their outputs with human intentions. My daily work in AI competitions requires the same, with agents and participants instead of rocks. The question is the same. How do you ensure local optimization aligns with the global objective?

What is Alignment?

Alignment is the problem of ensuring a system’s goals match the designer’s intentions. In AI, this breaks down into two layers. Inner alignment and outer alignment.

Inner alignment asks if the model’s internal goal match its training objective. Outer alignment asks if the training objectives match human values. The AI alignment literature frames this as the central challenge of AI safety. How to build systems that reliably do what we want, rather than merely what we ask for.

Community alignment faces the same core problem. Specifically, how do you get dozens or hundreds of participants to collectively solve a problem? They must avoid gaming the system or chasing local optima that harm the global objective. AI competitions tackle this with scoring systems that reward understanding over pattern matching.

The parallel is direct. AI alignment asks us to ensure a model’s optimization aligns with human values. While community alignment asks us to ensure participant optimization aligns with the competition’s objective. Both involve aggregating diverse inputs toward coherent outcomes.

The Alignment Framework

Alignment systems share common building blocks across domains:

1. Objective Functions: How we define success
2. Incentive Structures: How we motivate the right behaviors
3. Assessment Metrics: How we measure alignment
4. Feedback Loops: How systems learn and adapt

In AI, the loss function is the objective. In competitions, it is the scoring algorithm. In both cases, poorly designed objectives create perverse incentives.

A loss function that only rewards accuracy on a test set encourages memorization over generalization. A scoring system that only rewards leaderboard position encourages overfitting over robust methods.

The best systems use multi-dimensional assessment that cannot be gamed. Our most recent competition’s geometric mean scoring is a study in this principle. Excelling at one dimension cannot compensate for failing at another.

Case Study: No Second Guessing (Agentic)

No Second Guessing is the most recent expression of alignment thinking in our ecosystem. It was an agentic workflow competition for petrophysical interpretation, 800 wells (75 percent real, 25 percent honeypots), with the goal of 100 percent accuracy in identifying pay zones. Thanks to Oggy and Ashley for helping me create and organize this competition.

The Scoring Algorithm

The final score was the geometric mean of four equally weighted axes, each scored 0 to 100. The geometric mean formed the key design decision, forcing balanced performance. A zero on any axis zeros your entire score. Optimizing for one metric cannot game the system.

Axis 1: Gate Compliance

Each well faced 25 physics-based logic gates encoding hard limits, porosity ≤ 1.0, 0 ≤ water saturation ≤ 1.0, permeability ≥ 0. These limits are non-negotiable. Missing a curve results in automatic failure. The gates act as AI safety guardrails, hard boundaries that cannot be violated.

Axis 2: Pay Accuracy

Two components, equally weighted. - Footage error: How close is the total pay footage to the answer key? - Jaccard overlap: Are pay zones in the right places?

Get the total right but in the wrong locations means only half credit. The scoring prevented degenerate methods that match aggregate statistics without understanding spatial distribution.

Axis 3: Honeypot Rejection

25 percent of wells were honeypots containing baked-in physics violations. A correct workflow produced zero pay on these. The scoring applied a power penalty (caught/total)² × 100.

Missing 3 of 30 honeypots did not cost 10 points. It cost 19. Missing 10 drops the score to 44.4. The penalty scales quadratically because the system wanted zero tolerance for errors. The penalty embodies adversarial testing in action. Could agents distinguish true patterns from deliberate fakes?

Axis 4: Petrophysical Accuracy

On real wells, scoring compares three computed curves (effective porosity, water saturation, log-permeability) to the answer key within pay zones using RMSE, with generous tolerances matching human expert variability.

The geometric mean across the three curves ensured participants could not nail one while ignoring the others. Physics linked porosity, saturation, and permeability. The submissions respected those relationships.

How This Worked

No Second Guessing embedded multiple alignment systems:

1. Geometric mean forced balanced performance across all axes
2. Power penalty on honeypots made lazy methods mathematically impossible
3. Physics gates acted as hard limits that cannot be violated
4. Public/private leaderboard split prevented overfitting to visible data
5. Methodological restrictions (agentic workflows, no brute force via submission limits) enforced alignment with the intended approach

Participants built systems that understood petrophysics, beyond mere patterns in the data.

Alignment Across Problem Structures

My day-to-day work spans agentic, structured, semi-structured, and unstructured domains. The alignment systems adapt to each problem structure while maintaining the same core concepts.

Heat Signature Zero (Structured)

Heat Signature Zero presented a classic inverse problem. Identify heat sources from temperature time-series data on a 2D thermal plate. Thanks to Ashkan for building and releasing this competition.

The system combines accuracy (RMSE between predicted and actual sensor temperatures) with diversity of methods. λ=0.3 balanced the two. The essential requirement was that methods use the simulation engine during inference.

The system demonstrates methodological alignment in action. Participants could not brute-force grid search their way to the top.

The simulator acted as an oracle providing physics-grounded feedback. The ban on brute-force methods forces participants to participate in the intended process: iterative refinement using simulation feedback. The approach mirrors AI systems that must use interpretability tools or follow ethical limits during operation.

Reflection Connection (Semi-Structured)

Reflection Connection applied Siamese Neural Networks to seismic image similarity matching with only 8 classes and 500 training images. Thanks to Jakub for building this competition.

The system uses weighted accuracy on limited training categories. Confidence scores determine the impact on final score. The test checks whether models can generalize from limited examples to find similar patterns in a large corpus.

Here, alignment uses transfer learning. Taking a method from one domain and applying it to another. Transfer a computer vision approach to geophysics. The scoring rewarded models that understood seismic features, beyond mere memorization of the training set. Limited labels forced participants to build robust representations instead of overfitting.

Dynamic Rhythms (Unstructured)

Dynamic Rhythms asked participants to predict power outages correlated with rare extreme weather events using historical weather and outage datasets. Thanks to Ashkan for building this unconventional challenge.

The system uses multi-dimensional assessment that participants design themselves. They incorporated rare-event classification metrics, lead-time accuracy, severity predictions, and location accuracy metrics. The competition explicitly stated: “a good model should identify both whether and where an outage occurs, and also predict it sufficiently early to allow proactive measures.”

In this unstructured setting, alignment uses self-defined objectives that assess real-world utility. Participants designed metrics that prevented gaming and reward understanding of causal relationships between weather and outages.

The Pattern

Alignment systems scale with problem structure:

Common across all domains: each requires assessment that prevents gaming and rewards understanding.

The more structure a problem has, the more explicit the limits can be. The less structure, the more the assessment must focus on outcomes that matter in the real world.

What Works

Multi-dimensional metrics prevent gaming like geometric mean scoring (No Second Guessing) punishes imbalance more severely than arithmetic mean. Excelling at pay accuracy cannot compensate for failing at honeypot rejection.

Physics-based limits ground methods. The 25 physics gates in No Second Guessing are not arbitrary. They are fundamental laws. You cannot argue with thermodynamics or fluid dynamics. Hard limits that reflect real-world truth are the most reliable alignment system.

Adversarial testing reveals understanding. Honeypots are to AI what red teaming is to safety. They test whether your system has learned real patterns or just superficial correlations. The quadratic penalty ensures that distinction is not a minor consideration.

Methodological restrictions prevent degenerate methods. Banning brute force or requiring agentic workflows ensures participants participate in the intended problem-solving process. The restrictions embody process alignment: rewarding outcomes and how you achieve them.

Transparency builds trust. Clear assessment criteria, public leaderboards with holdout sets, and reproducible submission requirements create an environment where good methods rise to the top.

Parallels to AI Alignment

The parallels are striking.

• Multi-objective assessment: AI alignment needs more than accuracy on a test set. It needs safety, fairness, interpretability, and robustness metrics that cannot be traded off against each other.

• Limit satisfaction: Just as petrophysical interpretations must satisfy physics gates, AI systems must satisfy safety, ethical, and legal limits.

• Adversarial testing: Red teaming, honeypots, and adversarial examples reveal alignment failures that standard testing misses.

• Process matters: Interpretability, methodological soundness, and alignment with intended approaches are essential, beyond final performance numbers.

• Iterative refinement: Feedback loops allow systems to learn and adapt, improving alignment over time.

The Bigger Picture

My day-to-day work operates as a living laboratory for alignment problems. These problems test how to get dozens or hundreds of teams to collectively solve a problem without stepping on each other or gaming the system.

Community alignment asks, how do we get diverse participants to collectively solve a problem without local optima harming global objectives?

AI alignment asks, how do we get trillions of parameters to collectively serve human values?

Both require the same ingredients, clear objectives, good incentives, robust assessment, and iterative refinement.

The competitions prove alignment evolves. It changes with the problem. What works for structured problems (explicit limits) must adapt for unstructured ones (outcome-focused metrics).

The No Second Guessing ideal states it simply: good alignment means every action, at every scale, serves the goal.

In geology, that means the outcrop and the regional map tell the same story. In AI, every decision from model architecture to deployment serves human values. In community problem-solving, every participant’s optimization contributes to the collective objective.

Structured, semi-structured, unstructured, agentic. The alignment systems must adapt to the domain. The concepts remain constant. Define what you want, measure it honestly, prevent gaming, and reward understanding.

Closing

That formation pinchout in the desert taught me that alignment is not about being right at every scale. It is about maintaining coherence across scales.

The local outcrop observation must align with the regional understanding. The individual participant’s optimization must align with the community’s objective. The model’s outputs must align with human intentions.

The alignment journey never ends. For geologists, AI researchers, or communities. My work provides a testing ground for alignment ideas that scale from local optimization to global objectives.

The lessons are clear. Good alignment does not mean perfect agreement at every point. It means the outcrop and the map tell the same geologic story. The agent and the competition share the same objective. The AI and humanity pursue the same values.

And that, ultimately, is the point.

I am Jesse Pisel. Geologist, AI researcher, adventurer. By day I coax machines into predicting what lies beneath our feet. By night I coax my legs up mountains. I build tools, share the code, and leave some mysteries unanswered.

This site reflects those values. Simple by design, fast by choice, private by principle.

I grew up in the Western United States, where the land rewards directness and the sky stretches on forever. Years outdoors have taught me that resilient and maintainable beats disposable convenience every time. Running rivers has shown that the line matters more than force. Two decades between geology and AI have confirmed the same lesson that clarity beats clutter. A high signal-to-noise ratio is non-negotiable. The site extends that philosophy.

Design Choices

The design takes direct inspiration from Lowtech Magazine, whose solar-powered website runs on a balcony in Barcelona. It goes offline during cloudy weather. They consider this a feature. A vibrant online presence does not require high-energy infrastructure or perpetual availability. Sometimes the best statement is knowing when to go quiet.

The site also rejects the creeping homogeneity of the modern web. Corporate conformism has turned the internet into a strip mall. A/B testing and template-driven platforms have bleached the individuality out of sites like sun on cheap plastic. Different used to be a point of pride. Unique designs were badges of honor. The corporate web sandblasts that pride away, where every page looks like it was stamped from the same mold. The site remembers when the web had character, edges, and far less bloat.

According to Unlighthouse, fieldramblings.co before optimization weighed in at a svelte 855.4 kilobytes for mobile (sans this blog). It ranked in the top 25 percent of the lightest websites on the HTTP Archive scale. Only ten requests. Performance score of 74. The breakdown: 84 percent images, 15 percent comes from a single Tailwind script, and 1 percent fonts and markup. I am working to see how low that number can actually go.

What you will not find: analytics tracking your movements, consent banners cluttering the view, third-party scripts collecting your data, infinite scroll, lazy-loaded tracking pixels, or background calls to shadowy endpoints. The site does not phone home.

Lightweight pages load fast on slow connections. Rural broadband. Mobile data caps. Developing nations with limited infrastructure. Bloated sites leave people stranded in these places. Fast-loading pages address inequality. Not everyone has unlimited data or lives in a city with fiber. The web should work for everyone, not just the privileged few with fast connections.

The Design Philosophy

The modern web has a JavaScript problem. The average website now serves between 650 and 700 kilobytes of it with every page load. That is double the weight of 2015, nearly eight times 2010. The result? Fifty-three percent of users abandon sites taking more than three seconds to appear. A single second of delay costs seven percent in conversions, but as a blog I am not too concerned. It speaks to how quickly people bounce from slow pages. As of May 2026, only fifty-ish percent of sites pass all Core Web Vitals. JavaScript drags down performance worse than images or video, byte for byte. It is the rock your buddies slipped into your pack while you slept.

Surveillance plagues the modern web. The typical website arrives with an entourage of trackers, consent banners, analytics scripts, chat widgets, A/B testing tools. Browser fingerprinting silently collects signals to identify users without cookies such as screen resolution, installed fonts, timezone, browser version, hardware configuration. These signals create a unique digital silhouette. Privacy laws now require honoring browser signals. Most sites ignore them. The modern web has become a surveillance platform wrapped in the guise of convenience. I declined the invitation.

The Hosting Decision

I went hunting for a durable, low-surveillance home for this site. Looked into the indie hosts: NixiHost, PikaPods, AccuWeb, Hostinger, WebHostMost, Krystal. Turns out none met all four criteria of low cost, ease of use, solid support, and actually running on their own hardware. Most so-called indie hosts are just reselling Amazon or Google infrastructure. The emperor has no servers.

Cloudflare Pages emerged as the least evil option. Unlimited bandwidth for static assets. A free tier with no bandwidth caps and global distribution. Independent audits back their privacy commitments. They do not sell user data. They publish transparency reports. They have even challenged disproportionate censorship in court. Until something better comes along, this will do. I would rather poke myself in the eye than cut off a leg.

So here we are. A site that weighs 855 kilobytes. No trackers. No surveillance. No bloat. It loads fast on slow connections because not everyone has fiber. It respects privacy because not everyone wants to be watched.

These choices extend beyond technology. They are the same choices I make in life. Travel light, move with purpose, leave no trace. A fast site is an honest site. A private site is a respectful site. A simple site is a maintainable site.

This site works for everyone. That is the point.