[{"data":1,"prerenderedAt":1458},["ShallowReactive",2],{"navigation":3,"\u002Fnews\u002Finsights\u002Ftile-coding-real-time-prediction":357,"\u002Fnews\u002Finsights\u002Ftile-coding-real-time-prediction-surround":718},[4,8,17,21,25,29,33,345,349,353],{"title":5,"path":6,"stem":7},"About","\u002Fabout","about",{"title":9,"path":10,"stem":11,"children":12},"Authentication","\u002Fauth","auth",[13],{"title":14,"path":15,"stem":16},"Email Confirmation","\u002Fauth\u002Fconfirmation","auth\u002Fconfirmation",{"title":18,"path":19,"stem":20},"Industry Briefs","\u002Fcase-studies","case-studies",{"title":22,"path":23,"stem":24},"Contact","\u002Fcontact","contact",{"title":26,"path":27,"stem":28},"Thinkata | Production AI Engineering","\u002F","index",{"title":30,"path":31,"stem":32},"Insights","\u002Finsights","insights",{"title":34,"path":35,"stem":36,"children":37},"News","\u002Fnews","news",[38,41,65],{"title":39,"path":35,"stem":40},"News & Insights","news\u002Findex",{"title":18,"path":42,"stem":43,"children":44},"\u002Fnews\u002Fcase-studies","news\u002Fcase-studies",[45,49,53,57,61],{"title":46,"path":47,"stem":48},"Building Secure and Scalable AI Infrastructure: Integrating with Existing Systems through Modern Cloud Frameworks","\u002Fnews\u002Fcase-studies\u002Fcloud-infrastructure-ai","news\u002Fcase-studies\u002Fcloud-infrastructure-ai",{"title":50,"path":51,"stem":52},"Making Sense of Financial Regulations: How AI Teams Can Tackle Complex Documents","\u002Fnews\u002Fcase-studies\u002Ffinancial-regulations","news\u002Fcase-studies\u002Ffinancial-regulations",{"title":54,"path":55,"stem":56},"AI-Powered Transformations in Healthcare","\u002Fnews\u002Fcase-studies\u002Fhealth-care","news\u002Fcase-studies\u002Fhealth-care",{"title":58,"path":59,"stem":60},"Generative AI in Upstream Natural Gas: Shell's Exploration Initiative","\u002Fnews\u002Fcase-studies\u002Foil-gas","news\u002Fcase-studies\u002Foil-gas",{"title":62,"path":63,"stem":64},"Optimizing Manufacturing with AI-Driven Multi-Agent Systems","\u002Fnews\u002Fcase-studies\u002Fsupply-chain-optimization","news\u002Fcase-studies\u002Fsupply-chain-optimization",{"title":30,"path":66,"stem":67,"children":68},"\u002Fnews\u002Finsights","news\u002Finsights",[69,73,77,81,85,89,93,97,101,105,109,113,117,121,125,129,133,137,141,145,149,153,157,161,165,169,173,177,181,185,189,193,197,201,205,209,213,217,221,225,229,233,237,241,245,249,253,257,261,265,269,273,277,281,285,289,293,297,301,305,309,313,317,321,325,329,333,337,341],{"title":70,"path":71,"stem":72},"Nothing Gets Deleted, Just Blurred","\u002Fnews\u002Finsights\u002Fadaptive-context-memory","news\u002Finsights\u002Fadaptive-context-memory",{"title":74,"path":75,"stem":76},"The Capability-Reliability Split in Agent Systems","\u002Fnews\u002Finsights\u002Fagent-capability-reliability-split","news\u002Finsights\u002Fagent-capability-reliability-split",{"title":78,"path":79,"stem":80},"When Code Gets Cheap, Judgment Gets Expensive","\u002Fnews\u002Finsights\u002Fagentic-coding-engineer-skills","news\u002Finsights\u002Fagentic-coding-engineer-skills",{"title":82,"path":83,"stem":84},"The Rise of AI Agents in Cyberattacks: Latest Research and Threats","\u002Fnews\u002Finsights\u002Fai-agent-cyber-threats","news\u002Finsights\u002Fai-agent-cyber-threats",{"title":86,"path":87,"stem":88},"Winning the Bid Is Not the Same as Knowing the Price","\u002Fnews\u002Finsights\u002Fai-agent-economics","news\u002Finsights\u002Fai-agent-economics",{"title":90,"path":91,"stem":92},"The Smart Enterprise AI Stack: Why Teams of AI Agents Beat Solo Models Consistently","\u002Fnews\u002Finsights\u002Fai-architecture","news\u002Finsights\u002Fai-architecture",{"title":94,"path":95,"stem":96},"When Seeing Everything Becomes the Only Option","\u002Fnews\u002Finsights\u002Fai-comprehensive-observability","news\u002Finsights\u002Fai-comprehensive-observability",{"title":98,"path":99,"stem":100},"The Data Infrastructure AI-Native Systems Can't Ignore","\u002Fnews\u002Finsights\u002Fai-data-layer","news\u002Finsights\u002Fai-data-layer",{"title":102,"path":103,"stem":104},"Enterprise AI Triage Systems: Intelligent Automation for Large-Scale Operations","\u002Fnews\u002Finsights\u002Fai-enterprise-triage","news\u002Finsights\u002Fai-enterprise-triage",{"title":106,"path":107,"stem":108},"When Oversight Becomes Infrastructure","\u002Fnews\u002Finsights\u002Fai-governed-autonomy","news\u002Finsights\u002Fai-governed-autonomy",{"title":110,"path":111,"stem":112},"Designing for Graceful Failure in Compound AI Systems","\u002Fnews\u002Finsights\u002Fai-graceful-failure","news\u002Finsights\u002Fai-graceful-failure",{"title":114,"path":115,"stem":116},"Intelligent Composability: Building AI Systems Like Orchestra, Not Soloists","\u002Fnews\u002Finsights\u002Fai-intelligent-composability","news\u002Finsights\u002Fai-intelligent-composability",{"title":118,"path":119,"stem":120},"Building the Plane While Flying It — Migrating from Monolith to AI-Native Without Stopping","\u002Fnews\u002Finsights\u002Fai-migration-path","news\u002Finsights\u002Fai-migration-path",{"title":122,"path":123,"stem":124},"Stability Through Continuous Adaptation","\u002Fnews\u002Finsights\u002Fai-native-overview","news\u002Finsights\u002Fai-native-overview",{"title":126,"path":127,"stem":128},"Provable Stability: Mathematical Guarantees for Adaptive AI Systems","\u002Fnews\u002Finsights\u002Fai-provable-stability","news\u002Finsights\u002Fai-provable-stability",{"title":130,"path":131,"stem":132},"How Temperature Tuning Makes or Breaks Reinforcement Learning","\u002Fnews\u002Finsights\u002Fai-soft-actor-critic-entropy-collapse","news\u002Finsights\u002Fai-soft-actor-critic-entropy-collapse",{"title":134,"path":135,"stem":136},"Testing What Can't Be Predicted","\u002Fnews\u002Finsights\u002Fai-systems-testing","news\u002Finsights\u002Fai-systems-testing",{"title":138,"path":139,"stem":140},"Delegating the Job Is Not the Same as Delegating the Rules","\u002Fnews\u002Finsights\u002Fauthorization-drift-multi-agent","news\u002Finsights\u002Fauthorization-drift-multi-agent",{"title":142,"path":143,"stem":144},"Closing the Loop: How Human Corrections Can Make AI Systems Smarter Over Time","\u002Fnews\u002Finsights\u002Fclosing-the-loop","news\u002Finsights\u002Fclosing-the-loop",{"title":146,"path":147,"stem":148},"A Panel Can Be Safer Than Any Member of It","\u002Fnews\u002Finsights\u002Fcoalitional-alignment-panel-review","news\u002Finsights\u002Fcoalitional-alignment-panel-review",{"title":150,"path":151,"stem":152},"Multi-Path Reasoning: Collaborative and Competitive Approaches in AI","\u002Fnews\u002Finsights\u002Fcollaborative-competitive-agents","news\u002Finsights\u002Fcollaborative-competitive-agents",{"title":154,"path":155,"stem":156},"Why Challenges Supercharge Smarts for Humans and AI","\u002Fnews\u002Finsights\u002Fcompetition-improves-ai","news\u002Finsights\u002Fcompetition-improves-ai",{"title":158,"path":159,"stem":160},"Context is Infrastructure, Not Instructions","\u002Fnews\u002Finsights\u002Fcontext-is-infrastructure","news\u002Finsights\u002Fcontext-is-infrastructure",{"title":162,"path":163,"stem":164},"Context is the New Code","\u002Fnews\u002Finsights\u002Fcontext-is-new-code","news\u002Finsights\u002Fcontext-is-new-code",{"title":166,"path":167,"stem":168},"Continuous Thought Machines","\u002Fnews\u002Finsights\u002Fcontinuous-thought-machines","news\u002Finsights\u002Fcontinuous-thought-machines",{"title":170,"path":171,"stem":172},"Don't Vibe, Architect","\u002Fnews\u002Finsights\u002Fdont-vibe-architect","news\u002Finsights\u002Fdont-vibe-architect",{"title":174,"path":175,"stem":176},"The Edge of the Underdefined","\u002Fnews\u002Finsights\u002Fedge-of-the-underdefined","news\u002Finsights\u002Fedge-of-the-underdefined",{"title":178,"path":179,"stem":180},"Experts All the Way Down","\u002Fnews\u002Finsights\u002Fexperts-all-the-way","news\u002Finsights\u002Fexperts-all-the-way",{"title":182,"path":183,"stem":184},"A Multi-Tier Safety Architecture for Critical Applications","\u002Fnews\u002Finsights\u002Ffour-tier-architecture","news\u002Finsights\u002Ffour-tier-architecture",{"title":186,"path":187,"stem":188},"Green Dashboard, Unhappy Users","\u002Fnews\u002Finsights\u002Fgreen-dashboard-unhappy-users","news\u002Finsights\u002Fgreen-dashboard-unhappy-users",{"title":190,"path":191,"stem":192},"Hybrid Autoregressive Residual Tokens","\u002Fnews\u002Finsights\u002Fhart-model","news\u002Finsights\u002Fhart-model",{"title":194,"path":195,"stem":196},"Hierarchical Reasoning in Artificial Intelligence","\u002Fnews\u002Finsights\u002Fhierarchical-approaches","news\u002Finsights\u002Fhierarchical-approaches",{"title":198,"path":199,"stem":200},"Taking Turns Instead of Splitting the Difference","\u002Fnews\u002Finsights\u002Finterleaved-distillation-rl","news\u002Finsights\u002Finterleaved-distillation-rl",{"title":202,"path":203,"stem":204},"Latent Diffusion for Language Generation: A Comprehensive Overview","\u002Fnews\u002Finsights\u002Flatent-diffusion-for-language","news\u002Finsights\u002Flatent-diffusion-for-language",{"title":206,"path":207,"stem":208},"Breaking Language Barriers: How AI Can Translate Without Examples","\u002Fnews\u002Finsights\u002Flearning-languages","news\u002Finsights\u002Flearning-languages",{"title":210,"path":211,"stem":212},"The Emergence of AI Deception: How Large Language Models Have Learned to Strategically Mislead Users","\u002Fnews\u002Finsights\u002Fllm-deception","news\u002Finsights\u002Fllm-deception",{"title":214,"path":215,"stem":216},"Grading on a Shared Curve","\u002Fnews\u002Finsights\u002Fllm-judge-correlated-errors","news\u002Finsights\u002Fllm-judge-correlated-errors",{"title":218,"path":219,"stem":220},"Synergizing Specialized Reasoning and General Capabilities in AI","\u002Fnews\u002Finsights\u002Fllm-reasoning-advances","news\u002Finsights\u002Fllm-reasoning-advances",{"title":222,"path":223,"stem":224},"The Expensive Default","\u002Fnews\u002Finsights\u002Fllm-routing-cost-quality","news\u002Finsights\u002Fllm-routing-cost-quality",{"title":226,"path":227,"stem":228},"The AI That Rewrites Itself: MIT's Breakthrough in Self-Adapting Language Models","\u002Fnews\u002Finsights\u002Fllm-seal","news\u002Finsights\u002Fllm-seal",{"title":230,"path":231,"stem":232},"Forgetting on Request","\u002Fnews\u002Finsights\u002Fmachine-unlearning-removal-suppression","news\u002Finsights\u002Fmachine-unlearning-removal-suppression",{"title":234,"path":235,"stem":236},"The Map and the Walk","\u002Fnews\u002Finsights\u002Fmap-and-walk","news\u002Finsights\u002Fmap-and-walk",{"title":238,"path":239,"stem":240},"Metacognitive Reinforcement Learning for Self-Improving AI Systems","\u002Fnews\u002Finsights\u002Fmetacognitive-reinforcement-learning","news\u002Finsights\u002Fmetacognitive-reinforcement-learning",{"title":242,"path":243,"stem":244},"Revolutionary Advancements in Mixture of Experts (MoE) Architectures","\u002Fnews\u002Finsights\u002Fmixture-of-experts","news\u002Finsights\u002Fmixture-of-experts",{"title":246,"path":247,"stem":248},"One Model, Many Customers, and the Leak Nobody Tests For","\u002Fnews\u002Finsights\u002Fmulti-tenant-ai-isolation","news\u002Finsights\u002Fmulti-tenant-ai-isolation",{"title":250,"path":251,"stem":252},"Balancing Neural Plasticity and Stability","\u002Fnews\u002Finsights\u002Fneural-plasticity","news\u002Finsights\u002Fneural-plasticity",{"title":254,"path":255,"stem":256},"Offline RL and the Data Flywheel","\u002Fnews\u002Finsights\u002Foffline-rl-data-flywheel","news\u002Finsights\u002Foffline-rl-data-flywheel",{"title":258,"path":259,"stem":260},"Orchestration Without a Signal From the Room","\u002Fnews\u002Finsights\u002Fplanner-in-the-loop","news\u002Finsights\u002Fplanner-in-the-loop",{"title":262,"path":263,"stem":264},"Second-Guessing Has a Price","\u002Fnews\u002Finsights\u002Freasoning-budget-allocation","news\u002Finsights\u002Freasoning-budget-allocation",{"title":266,"path":267,"stem":268},"Reasoning You Can Check","\u002Fnews\u002Finsights\u002Freasoning-you-can-check","news\u002Finsights\u002Freasoning-you-can-check",{"title":270,"path":271,"stem":272},"When Optimization Optimizes Itself","\u002Fnews\u002Finsights\u002Frecursive-goodhart","news\u002Finsights\u002Frecursive-goodhart",{"title":274,"path":275,"stem":276},"Reward Design as Architecture","\u002Fnews\u002Finsights\u002Freward-design-as-architecture","news\u002Finsights\u002Freward-design-as-architecture",{"title":278,"path":279,"stem":280},"When Success Has No Author: The Temporal Credit Assignment Problem","\u002Fnews\u002Finsights\u002Frl-credit-assignment-problem","news\u002Finsights\u002Frl-credit-assignment-problem",{"title":282,"path":283,"stem":284},"Beyond Entropy Collapse: When Exploration Succeeds but Learning Fails","\u002Fnews\u002Finsights\u002Frl-optimization-gaps","news\u002Finsights\u002Frl-optimization-gaps",{"title":286,"path":287,"stem":288},"The Body Was Supposed to Be the Easy Part","\u002Fnews\u002Finsights\u002Frobot-embodiment-interface","news\u002Finsights\u002Frobot-embodiment-interface",{"title":290,"path":291,"stem":292},"The Path to Practical Confidential Computing for AI Systems","\u002Fnews\u002Finsights\u002Fsecure-ai-architectures","news\u002Finsights\u002Fsecure-ai-architectures",{"title":294,"path":295,"stem":296},"When an AI Edits Its Own Memory","\u002Fnews\u002Finsights\u002Fself-editing-agent-memory","news\u002Finsights\u002Fself-editing-agent-memory",{"title":298,"path":299,"stem":300},"The Guard at the Front Door","\u002Fnews\u002Finsights\u002Fshallow-safety-alignment-refusal","news\u002Finsights\u002Fshallow-safety-alignment-refusal",{"title":302,"path":303,"stem":304},"Guess First, Check Later","\u002Fnews\u002Finsights\u002Fspeculative-execution-pattern","news\u002Finsights\u002Fspeculative-execution-pattern",{"title":306,"path":307,"stem":308},"Spiking Neural Networks for Energy-Efficient AI","\u002Fnews\u002Finsights\u002Fspiking-neural-networks","news\u002Finsights\u002Fspiking-neural-networks",{"title":310,"path":311,"stem":312},"When Replay Is Not an Option: Streaming Q-Learning and SARSA Get a Second Look","\u002Fnews\u002Finsights\u002Fstreaming-q-learning-revival","news\u002Finsights\u002Fstreaming-q-learning-revival",{"title":314,"path":315,"stem":316},"The Turn as the Unit of Quality","\u002Fnews\u002Finsights\u002Fstructured-iteration-quality","news\u002Finsights\u002Fstructured-iteration-quality",{"title":318,"path":319,"stem":320},"AI Speech Translation: Breaking Down Language Barriers","\u002Fnews\u002Finsights\u002Fsts-performance-advances","news\u002Finsights\u002Fsts-performance-advances",{"title":322,"path":323,"stem":324},"Test-Time Training Layers: The Next Evolution in Transformer Architecture","\u002Fnews\u002Finsights\u002Ftest-time-training-layers","news\u002Finsights\u002Ftest-time-training-layers",{"title":326,"path":327,"stem":328},"The Part You Meant to Keep","\u002Fnews\u002Finsights\u002Fthe-part-you-meant-to-keep","news\u002Finsights\u002Fthe-part-you-meant-to-keep",{"title":330,"path":331,"stem":332},"Predicting the Next Second: Tile Coding as a Real-Time Prediction Engine","\u002Fnews\u002Finsights\u002Ftile-coding-real-time-prediction","news\u002Finsights\u002Ftile-coding-real-time-prediction",{"title":334,"path":335,"stem":336},"Breakthrough: Large Language Models Pass the Turing Test","\u002Fnews\u002Finsights\u002Fturing-tests","news\u002Finsights\u002Fturing-tests",{"title":338,"path":339,"stem":340},"Algorithms Used in Autonomous Fighter Jet Flight Research","\u002Fnews\u002Finsights\u002Fvenom-f16-flight-autonomy","news\u002Finsights\u002Fvenom-f16-flight-autonomy",{"title":342,"path":343,"stem":344},"Training in a World That Does Not Exist Yet","\u002Fnews\u002Finsights\u002Fworld-models-as-infrastructure","news\u002Finsights\u002Fworld-models-as-infrastructure",{"title":346,"path":347,"stem":348},"Privacy Policy","\u002Fprivacy","privacy",{"title":350,"path":351,"stem":352},"Research","\u002Fresearch","research",{"title":354,"path":355,"stem":356},"Terms of Service","\u002Fterms","terms",{"id":358,"title":330,"body":359,"date":699,"description":700,"extension":701,"image":702,"meta":703,"navigation":715,"path":331,"seo":716,"stem":332,"__hash__":717},"insights\u002Fnews\u002Finsights\u002Ftile-coding-real-time-prediction.md",{"type":360,"value":361,"toc":683},"minimark",[362,382,392,405,415,419,422,432,439,447,451,459,462,484,488,496,504,523,527,534,551,555,562,577],[363,364,367,368,367,375],"div",{"className":365},[366],"page-title","\n  ",[369,370,374],"h1",{"className":371,"id":373},[372],"page-title__main","predicting-the-next-second","Predicting the Next Second",[376,377,381],"h2",{"className":378,"id":380},[379],"page-title__sub","tile-coding-as-a-real-time-prediction-engine","Tile Coding as a Real-Time Prediction Engine",[363,383,385,386],{"style":384},"width: 100%; padding: 2%;","\n    ",[387,388],"img",{"src":389,"alt":390,"style":391},"https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1531259960912-746d0c872ee6?w=1200&auto=format&fit=crop","A cyclist in motion blur on a city street","width: 100%; height: auto;",[393,394,395,396,404],"p",{},"A cyclist weaving through traffic is not reacting to the street as it is. The body is already braced for the pothole a second ahead, already leaning into a turn that has not started yet. Psychologists call this constant, small-scale forecasting nexting, and a group of reinforcement learning researchers borrowed the word to describe a robot that learned to do something similar ",[397,398,399],"sup",{},[400,401,403],"a",{"href":402},"#source-1","[1]",". Reinforcement learning is the branch of machine learning in which software improves through trial and error, and one of its quieter strengths turns out to be learning to predict.",[393,406,407,408,414],{},"What made the robot practical was not a large neural network. It was tile coding, a simple technique with roots in an older control method known as CMAC ",[397,409,410],{},[400,411,413],{"href":412},"#source-2","[2]",", and one that most students meet briefly before moving on to deep learning. Looking at where tile coding has actually been put to work suggests it deserves a second look, less as a way to make machines act and more as a way to help them anticipate.",[376,416,418],{"id":417},"overlapping-grids","Overlapping Grids",[393,420,421],{},"Imagine laying a sheet of graph paper over a map. Any location now falls in one square, and everything learned about that square applies to the whole of it. That is a coarse way to describe a place, since two points just either side of a line end up in different squares and share nothing.",[393,423,424,425,431],{},"Now add a second sheet, shifted a little, and a third shifted a little more. A location falls in one square on every sheet. Two nearby locations share most of their squares, two distant ones share none, and the overlap between sheets lets the system tell apart points much finer than any single square. That is tile coding. Each sheet is a tiling, each square a tile, and the set of tiles a reading lands in becomes its description ",[397,426,427],{},[400,428,430],{"href":429},"#source-3","[3]",".",[393,433,434,435,431],{},"The same trick works for any measurement, not just position on a map. A motor's current, a joint's angle or a room's temperature can all be described this way. Richard Sutton's reference software for the technique, called tiles3, is short enough to fit on a single page ",[397,436,437],{},[400,438,413],{"href":412},[393,440,441,442,446],{},"From an engineering point of view, the appeal is predictability. Every reading lights up the same small number of tiles, so each learning step does the same amount of work, and the software caps how much memory the tiles can use ",[397,443,444],{},[400,445,413],{"href":412},". Learning about one situation also leaves unrelated situations untouched, which matters a great deal for a system that has to keep learning while it runs.",[376,448,450],{"id":449},"asking-questions-about-the-future","Asking Questions About the Future",[393,452,453,454,458],{},"The second idea is about what to predict. Reinforcement learning usually estimates how much reward lies ahead. The nexting research kept that machinery but pointed it at ordinary sensor readings instead, so the robot could learn how bright its surroundings were about to get, or how hard a motor would be working over the next few seconds ",[397,455,456],{},[400,457,403],{"href":402},". A single setting controls how far ahead each question looks, from a fraction of a second to several seconds.",[393,460,461],{},"Because every question reads from the same tiles, asking one more costs very little. There is no new model to build, only one more small set of numbers to keep up to date.",[363,463,385,465,385,469,385,475],{"style":464},"width: 100%; margin: 20px 0;",[387,466],{"src":467,"alt":468,"style":391},"https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1678225867994-e7a5b071ebfd?w=1200&auto=format&fit=crop","A small hobby robot with exposed wiring and an ultrasonic sensor",[470,471,474],"h3",{"style":472,"id":473},"margin: 1rem 0 0.5rem 0;","thousands-of-forecasts-on-a-laptop","Thousands of Forecasts on a Laptop",[393,476,478,479,483],{"style":477},"margin: 0;","Every wire on a hobby robot like this one carries a signal that could be forecast, battery level, wheel effort, the distance reported by the sensor up front. The research robot in the nexting study had several dozen such signals. Its learning system made more than two thousand forecasts about them at once, from an ordinary laptop, and still finished each round of updates comfortably inside the robot's tenth-of-a-second rhythm ",[397,480,481],{},[400,482,403],{"href":402},". Most of the learning was done within about half an hour of driving around its pen.",[376,485,487],{"id":486},"what-the-robot-learned","What the Robot Learned",[393,489,490,491,495],{},"Speed alone would not make the result interesting. The researchers also compared the robot's forecasts, learned on the fly, against the best forecasts that could be worked out afterward with all the data in hand, and the live versions held up well ",[397,492,493],{},[400,494,403],{"href":402},". A single set of tiles, built once, was enough to forecast light, motion and motor signals across every time horizon tested. Nobody had to design separate inputs for each kind of signal.",[393,497,498,499,503],{},"One small detail in the paper hints at a practical use. The robot's motors had overheating protection and would stop to cool down every so often ",[397,500,501],{},[400,502,403],{"href":402},". In the experiment that was just a nuisance. For anyone responsible for motors, pumps or fans, a forecast of the next few seconds of effort and heat, learned from nothing but the machine's own sensors, starts to look like an early warning.",[363,505,385,506,385,510,385,514],{"style":464},[387,507],{"src":508,"alt":509,"style":391},"https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1581090464777-f3220bbe1b8b?w=1200&h=700&auto=format&fit=crop","A robotic prosthetic hand raised against a teal background",[470,511,513],{"style":472,"id":512},"anticipating-the-next-joint","Anticipating the Next Joint",[393,515,516,517,431],{"style":477},"A powered hand like this one has more joints than its wearer has muscle signals to drive them. Many prosthetic arms controlled by muscle signals solve this with a switch that steps control from elbow to wrist to hand in a fixed order, an approach widely regarded as slow and difficult to use ",[397,518,519],{},[400,520,522],{"href":521},"#source-4","[4]",[376,524,526],{"id":525},"predictions-that-help-a-person","Predictions That Help a Person",[393,528,529,530,431],{},"Patrick Pilarski and colleagues at the University of Alberta asked whether the same kind of forecasting could reorder that list as a person used the arm, offering first whichever joint the person was most likely to want next. Their system tile coded the arm's position and other signals, learned its forecasts in real time on a laptop, and on a standard skills test beat even the best fixed order, one identified afterward from the same session, cutting the expected wait between joints by more than 14 percent ",[397,531,532],{},[400,533,522],{"href":521},[393,535,536,537,543,544,550],{},"Later work moved closer to everyday use. Real-time prediction learning of this kind was demonstrated with an amputee using a training robot arm ",[397,538,539],{},[400,540,542],{"href":541},"#source-5","[5]",", and a follow-up case series found that adaptive switching reduced both how often and how long people spent switching, for an amputee and for able-bodied participants ",[397,545,546],{},[400,547,549],{"href":548},"#source-6","[6]",". These are small studies, and clinicians are far better placed to judge what they mean for people who wear prostheses every day. From a machine learning perspective, the notable part is that the system learned from each person's own signals while the arm was in use, with no separate training phase beforehand.",[376,552,554],{"id":553},"what-this-suggests-for-builders","What This Suggests for Builders",[393,556,557,558,431],{},"For anyone building software for physical equipment, this work points to a pattern that seems underused. A handful of sensor readings, overlapping grids, and a set of simple forecasting questions can give a machine a running sense of what is about to happen, on hardware no bigger than a laptop and at a cost that is known in advance. The tiles3 manual covers most of the practical setup, such as choosing how wide each tile should be and which measurements to combine ",[397,559,560],{},[400,561,413],{"href":412},[393,563,564,565,569,570,576],{},"The catch is that the number of tiles explodes as more measurements are combined. The nexting robot, for instance, tiled most of its sensors on their own or in small pairs rather than all together ",[397,566,567],{},[400,568,403],{"href":402},". Deciding which measurements belong together is where knowledge of the equipment enters the design. For richer, more complex data, one research team working on laser welding first used a neural network to boil the raw data down to a few key features, then handed those to the same kind of lightweight forecasting ",[397,571,572],{},[400,573,575],{"href":574},"#source-7","[7]",". Whether that pairing, a learned summary in front of a cheap and predictable forecaster, becomes a common recipe for small devices is still an open question. It looks like a reasonable bet for machines that need to keep learning from their own sensors without ever stopping to retrain.",[363,578,367,582,367,585],{"className":579},[580,581],"references","mt-8",[376,583,584],{"id":580},"References",[586,587,385,593,385,611,385,623,385,635,385,647,385,659,385,671,367],"ol",{"className":588},[589,590,591,592],"list-decimal","list-inside","space-y-2","mt-4",[594,595,597,598,602,603],"li",{"id":596},"source-1","J. Modayil et al., \"Multi-timescale Nexting in a Reinforcement Learning Robot,\" in ",[599,600,601],"em",{},"From Animals to Animats 12 (SAB 2012), Lecture Notes in Computer Science",", vol. 7426, Springer, 2012, pp. 299–309. DOI: ",[400,604,610],{"href":605,"target":606,"className":607},"https:\u002F\u002Fdoi.org\u002F10.1007\u002F978-3-642-33093-3_30","_blank",[608,609],"text-blue-600","underline","[Online]",[594,612,614,615,618,619],{"id":613},"source-2","R. S. Sutton, \"Tile Coding Software, Reference Manual, Version 3.0,\" ",[599,616,617],{},"Reinforcement Learning and Artificial Intelligence Lab, University of Alberta",", accessed Nov. 6, 2026, ",[400,620,610],{"href":621,"target":606,"className":622},"http:\u002F\u002Fincompleteideas.net\u002Ftiles\u002Ftiles3.html",[608,609],[594,624,626,627,630,631],{"id":625},"source-3","R. S. Sutton and A. G. Barto, ",[599,628,629],{},"Reinforcement Learning: An Introduction",", 2nd ed., Cambridge, MA, USA, MIT Press, 2018, sec. 9.5.4, p. 217. ",[400,632,610],{"href":633,"target":606,"className":634},"http:\u002F\u002Fincompleteideas.net\u002Fbook\u002FRLbook2018.pdf#page=239",[608,609],[594,636,638,639,642,643],{"id":637},"source-4","P. M. Pilarski et al., \"Dynamic switching and real-time machine learning for improved human control of assistive biomedical robots,\" in ",[599,640,641],{},"Proc. 4th IEEE RAS\u002FEMBS International Conference on Biomedical Robotics and Biomechatronics (BioRob'12)",", Rome, Italy, 2012, pp. 296–302. DOI: ",[400,644,610],{"href":645,"target":606,"className":646},"https:\u002F\u002Fdoi.org\u002F10.1109\u002FBioRob.2012.6290309",[608,609],[594,648,650,651,654,655],{"id":649},"source-5","P. M. Pilarski et al., \"Adaptive artificial limbs: a real-time approach to prediction and anticipation,\" ",[599,652,653],{},"IEEE Robotics & Automation Magazine",", vol. 20, no. 1, pp. 53–64, 2013. DOI: ",[400,656,610],{"href":657,"target":606,"className":658},"https:\u002F\u002Fdoi.org\u002F10.1109\u002FMRA.2012.2229948",[608,609],[594,660,662,663,666,667],{"id":661},"source-6","A. L. Edwards et al., \"Application of real-time machine learning to myoelectric prosthesis control: A case series in adaptive switching,\" ",[599,664,665],{},"Prosthetics and Orthotics International",", vol. 40, no. 5, pp. 573–581, 2016. DOI: ",[400,668,610],{"href":669,"target":606,"className":670},"https:\u002F\u002Fdoi.org\u002F10.1177\u002F0309364615605373",[608,609],[594,672,674,675,678,679],{"id":673},"source-7","J. Günther et al., \"Intelligent laser welding through representation, prediction, and control learning: An architecture with deep neural networks and reinforcement learning,\" ",[599,676,677],{},"Mechatronics",", vol. 34, pp. 1–11, 2016. DOI: ",[400,680,610],{"href":681,"target":606,"className":682},"https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.mechatronics.2015.09.004",[608,609],{"title":684,"searchDepth":685,"depth":685,"links":686},"",2,[687,688,689,693,696,697,698],{"id":380,"depth":685,"text":381},{"id":417,"depth":685,"text":418},{"id":449,"depth":685,"text":450,"children":690},[691],{"id":473,"depth":692,"text":474},3,{"id":486,"depth":685,"text":487,"children":694},[695],{"id":512,"depth":692,"text":513},{"id":525,"depth":685,"text":526},{"id":553,"depth":685,"text":554},{"id":580,"depth":685,"text":584},"2026-11-06","Tile coding is usually taught as a classroom exercise in reinforcement learning. Its most practical use may be prediction, helping a robot or a prosthetic arm learn to anticipate what happens next while it runs, on modest hardware and with no separate training phase.","md",{"src":389},{"authors":704,"badge":710,"source":712},[705],{"avatar":706,"name":708,"to":709},{"src":707},"\u002Fimg\u002Fmark_avatar.png","Mark Williams","#",{"label":711},"Reinforcement Learning",{"name":713,"url":714},"Thinkata Research","https:\u002F\u002Fthinkata.com",true,{"title":330,"description":700},"aBKbRiZDaa0itNJxwsVAVsbUsxF-1fiWgFR524wtojw",[719,1010],{"id":720,"title":298,"body":721,"date":998,"description":999,"extension":701,"image":1000,"meta":1001,"navigation":715,"path":299,"seo":1008,"stem":300,"__hash__":1009,"_path":299},"insights\u002Fnews\u002Finsights\u002Fshallow-safety-alignment-refusal.md",{"type":360,"value":722,"toc":987},[723,735,741,744,747,754,758,766,769,772,780,784,792,800,803,807,819,827,845,858,861,865,875,878,882,893,896],[363,724,367,726,367,730],{"className":725},[366],[369,727,298],{"className":728,"id":729},[372],"the-guard-at-the-front-door",[376,731,734],{"className":732,"id":733},[379],"why-a-language-model-that-refuses-can-usually-still-comply","Why a Language Model That Refuses Can Usually Still Comply",[363,736,385,737],{"style":384},[387,738],{"src":739,"alt":740,"style":391},"https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1758448721205-8465cebc26af?w=1200&h=750&fit=crop&auto=format","A modern office lobby with a reception desk and elevators behind it",[393,742,743],{},"A model that answers \"I can't help with that\" sounds like a model that can't. Ask a chat assistant how to make a nerve agent or write working ransomware, and the polite refusal reads as a statement about what the model is able to do. Research from the past three years suggests it is usually a statement about what the model has been trained to say first.",[393,745,746],{},"The lobby of an office tower is a useful picture. A guard at the front desk checks badges and turns visitors away, and to anyone standing at the entrance the building looks closed. The offices upstairs are still full. Someone who comes in through the loading dock, or who is already past the desk, finds every floor unguarded. Safety training, the stage where a model learns to decline harmful requests, appears to work a lot like that guard. It is very good at the front door and much weaker everywhere else.",[393,748,749,750,753],{},"An earlier Thinkata piece on ",[400,751,752],{"href":231},"machine unlearning"," found that methods meant to make a model forget mostly make it quieter. Refusal raises the same question from the other side. A model that declines might be unable to help, or it might just be unwilling at that moment. The difference matters a great deal to anyone deciding how much to trust it.",[376,755,757],{"id":756},"a-guard-who-only-watches-the-door","A Guard Who Only Watches the Door",[393,759,760,761,765],{},"A language model writes its reply one token at a time. A token is a word or a piece of a word, and each new one is chosen based on everything written so far. To find where safety training actually changes that process, a team from Princeton and Google DeepMind compared a safety-trained model with the base model it started from ",[397,762,763],{},[400,764,403],{"href":402},". The differences were concentrated in the first few tokens of the reply. After that, the two models behaved much alike.",[393,767,768],{},"One experiment in the paper makes the point vividly. The researchers took a base model that had never had any safety training and simply forced its reply to begin with the words \"I cannot.\" That one change took it from giving harmful answers most of the time to almost never, close to the safety-trained version. A large share of what looks like learned judgment, in other words, can be reproduced by getting the opening words right.",[393,770,771],{},"The reverse works too. In a prefilling attack, the attacker writes the first few words of the reply, something like \"Sure, here is how,\" and lets the model continue. Some model providers allow this directly, and anyone running an open model can do it trivially. The more harmful words were placed at the start, the more often the safety-trained model carried on as if it had agreed. The paper received an outstanding paper award at ICLR 2025. It calls the pattern shallow safety alignment, and much of the follow-up work has adopted the term.",[393,773,774,775,779],{},"Why would training settle on the doorway? A 2026 preprint from the University of Southampton offers an explanation rooted in how these models generate text ",[397,776,777],{},[400,778,413],{"href":412},". Once a reply has started down a path, the model strongly tends to keep going along it, a property the authors call autoregressive consistency. A refusal that opens with \"I'm sorry, but\" almost writes itself from there. Training has little left to correct in the later words, so its changes land on the opening ones. On this account, the shallowness is less an accident of particular datasets than a side effect of next-word prediction itself. The argument rests on a theorem with stated assumptions and on tests of a few open models, and it has not yet been peer reviewed.",[376,781,783],{"id":782},"one-switch-on-the-wall","One Switch on the Wall",[393,785,786,787,791],{},"If refusal mostly lives in the opening words, what decides which opening the model picks? Researchers looking inside thirteen open chat models, the largest with 72 billion parameters, found what looks like a single control ",[397,788,789],{},[400,790,430],{"href":429},". Deep in a model's internal activations, the running numerical state it carries while processing a prompt, there was one direction associated with refusing. Erasing that direction left the models answering harmful requests they had previously declined. Adding it made them refuse perfectly harmless ones. Their other abilities came through largely intact.",[393,793,794,795,799],{},"A separate study arrives at a compatible picture by removing pieces of a model rather than steering it ",[397,796,797],{},[400,798,522],{"href":521},". The parts that matter most for safety turned out to be small and separate from the parts that make the model useful. They made up around 3 percent of its parameters, the adjustable numbers a model learns during training. Remove them and safety breaks down while the model's usefulness barely changes.",[393,801,802],{},"In the lobby picture, the guard answers to a single switch on the wall, and the switch is not wired to the offices upstairs. That separation is what makes the arrangement fragile. A capability and the refusal to use it are stored in different places, so it is possible to touch one without disturbing the other.",[376,804,806],{"id":805},"retraining-the-guard","Retraining the Guard",[393,808,809,810,814,815,431],{},"Most people will never edit a model's internal activations. Many more can fine-tune one, which means giving it a short round of extra training on examples of their choosing, a service several model providers sell. In 2023, researchers made a safety-trained model comply with nearly any harmful instruction using just ten carefully written examples ",[397,811,812],{},[400,813,542],{"href":541},". Fine-tuning on them through a commercial service cost under twenty cents. More troubling for anyone offering fine-tuning as a product, training on ordinary, harmless datasets also eroded safety, less dramatically but noticeably, even when nobody intended it. The shallow alignment paper suggests a reason. Ordinary training on benign data tends to move the opening tokens most, which is exactly where the guard stands ",[397,816,817],{},[400,818,403],{"href":402},[393,820,821,822,826],{},"An earlier analysis framed the deeper issue as mismatched generalization ",[397,823,824],{},[400,825,549],{"href":548},". A model's abilities reach widely, to unfamiliar languages, encoded text, and odd phrasings, because pretraining on vast amounts of text taught them broadly. Safety training covers a much narrower range of examples. Wherever the two diverge, the ability is present and the refusal is not. The authors argued that safety measures would need to be roughly as capable as the model they protect, and that scale alone would not close the gap.",[363,828,385,829,385,833,385,837],{"style":464},[387,830],{"src":831,"alt":832,"style":391},"https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1760484376470-5e66c195978b?w=1200&auto=format&fit=crop","A long empty hallway with doors on both sides",[470,834,836],{"style":472,"id":835},"guards-on-every-floor","Guards on Every Floor",[393,838,839,840,844],{"style":477},"Every door along a corridor is another place to slip in, and a guard posted only in the lobby never sees most of them. The obvious repair is to train the model to recover partway through a reply, not just at the start. The shallow alignment paper did this with what it calls safety recovery examples. These are replies that begin harmfully and then turn back to a refusal. Training on them made prefilling and several other attacks far less effective ",[397,841,842],{},[400,843,403],{"href":402},". The same paper proposed a training rule that limits how far later fine-tuning can move the first few tokens, guarding the doorway itself.",[393,846,847,848,852,853,857],{},"Follow-up work suggests the corridor is longer than it first appeared. The Southampton group's random insertion attack drops a short harmful passage into the middle of an otherwise firm refusal. Models often pick up the harmful thread and continue it, even after refusing for a long stretch ",[397,849,850],{},[400,851,413],{"href":412},". A second 2026 preprint, from researchers in Seoul, found that injecting a few words at almost any point in a reply could change what the model did next ",[397,854,855],{},[400,856,575],{"href":574},". The effect was strongest early on, but no point in the reply was fully safe. Both groups trained models on deliberately interrupted replies and reported large drops in attack success.",[393,859,860],{},"The Seoul study adds a complication. Models whose internal state lined up strongly with the refusal direction were not, as a result, any harder to push off course. A model can look internally committed to refusing and still give in. The single switch may be a handy lever for attackers, then, without being a reliable gauge for defenders. The deeper training also made two of the models turn down harmless requests noticeably more often, a reminder that a guard on every floor costs something in friction.",[376,862,864],{"id":863},"locking-the-rooms","Locking the Rooms",[393,866,867,868,874],{},"A different line of defense leaves the refusal alone and acts on the capability. Circuit breakers, introduced in 2024, watch for the internal patterns a model produces when it starts generating harmful content, and scramble them as they appear ",[397,869,870],{},[400,871,873],{"href":872},"#source-8","[8]",". The model loses its train of thought instead of declining politely. Because the method targets what the model is doing rather than how its reply begins, a prefilled \"Sure, here is\" does not get around it in the same way. The authors report strong resistance to attacks the system was never trained against, without much loss in usefulness. In the lobby picture, this is closer to locking the rooms upstairs than to adding guards.",[393,876,877],{},"Locked rooms are still full rooms. The knowledge remains in the model, and someone with access to the weights could in principle retrain the lock away. That leads back to the unlearning question. A model that truly cannot help would have to lack the ability, and current methods struggle to show they have removed anything at all.",[376,879,881],{"id":880},"what-a-refusal-can-honestly-claim","What a Refusal Can Honestly Claim",[393,883,884,885,888,889,892],{},"From a systems perspective, the evidence points toward treating a refusal as a behavior observed under particular conditions, rather than as a fixed property of the model. A claim that a model declines harmful requests is only as strong as the list of access routes it was tested against. A closed model behind an interface that blocks prefilled replies and custom fine-tuning faces far fewer attacks than the same model with those features turned on. An open model, whose weights anyone can download, faces all of them, including direct removal of the refusal direction. Decisions about which features to expose are, in effect, safety decisions. Earlier Thinkata pieces on ",[400,886,887],{"href":107},"enforcement that sits outside the model"," and on ",[400,890,891],{"href":247},"isolation between customers"," make a related case for controls that do not depend on the model's own good behavior.",[393,894,895],{},"The harder question is how anyone could tell a model that will not from a model that cannot. The closest available test is to deliberately try to undo a model's safety training and then check what it can still do. Like the relearning attacks in unlearning research, that only sets a floor, since a more determined attacker might recover more. Specialists in biosecurity or cybersecurity would know better than a systems reading can how much leftover ability actually matters in their fields. The research so far does make two things fairly clear. The guard at the front desk is doing real work, and the building behind the guard is still full.",[363,897,367,899,367,901],{"className":898},[580,581],[376,900,584],{"id":580},[586,902,385,904,385,915,385,926,385,937,385,948,385,958,385,968,385,977,367],{"className":903},[589,590,591,592],[594,905,906,907,910,911],{"id":596},"X. Qi et al., \"Safety Alignment Should Be Made More Than Just a Few Tokens Deep,\" in ",[599,908,909],{},"Proc. International Conference on Learning Representations (ICLR'25)",", 2025, ",[400,912,610],{"href":913,"target":606,"className":914},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.05946",[608,609],[594,916,917,918,921,922],{"id":613},"B. Lyu et al., \"When Autoregressive Consistency Hurts Safety Alignment,\" ",[599,919,920],{},"arXiv",", 2026, ",[400,923,610],{"href":924,"target":606,"className":925},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.04168",[608,609],[594,927,928,929,932,933],{"id":625},"A. Arditi et al., \"Refusal in Language Models Is Mediated by a Single Direction,\" in ",[599,930,931],{},"Advances in Neural Information Processing Systems",", vol. 37, 2024, ",[400,934,610],{"href":935,"target":606,"className":936},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.11717",[608,609],[594,938,939,940,943,944],{"id":637},"B. Wei et al., \"Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications,\" in ",[599,941,942],{},"Proc. 41st International Conference on Machine Learning (ICML'24)",", 2024, ",[400,945,610],{"href":946,"target":606,"className":947},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.05162",[608,609],[594,949,950,951,943,954],{"id":649},"X. Qi et al., \"Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!,\" in ",[599,952,953],{},"Proc. International Conference on Learning Representations (ICLR'24)",[400,955,610],{"href":956,"target":606,"className":957},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.03693",[608,609],[594,959,960,961,963,964],{"id":661},"A. Wei, N. Haghtalab, and J. Steinhardt, \"Jailbroken: How Does LLM Safety Training Fail?,\" in ",[599,962,931],{},", vol. 36, 2023, ",[400,965,610],{"href":966,"target":606,"className":967},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.02483",[608,609],[594,969,970,971,921,973],{"id":673},"K. Park and T. Kim, \"Inference-Time Vulnerability Beyond Shallow Safety: Alignment Along Generation Trajectories,\" ",[599,972,920],{},[400,974,610],{"href":975,"target":606,"className":976},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.04778",[608,609],[594,978,980,981,932,983],{"id":979},"source-8","A. Zou et al., \"Improving Alignment and Robustness with Circuit Breakers,\" in ",[599,982,931],{},[400,984,610],{"href":985,"target":606,"className":986},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.04313",[608,609],{"title":684,"searchDepth":685,"depth":685,"links":988},[989,990,991,992,995,996,997],{"id":733,"depth":685,"text":734},{"id":756,"depth":685,"text":757},{"id":782,"depth":685,"text":783},{"id":805,"depth":685,"text":806,"children":993},[994],{"id":835,"depth":692,"text":836},{"id":863,"depth":685,"text":864},{"id":880,"depth":685,"text":881},{"id":580,"depth":685,"text":584},"2026-11-13","A language model that declines a harmful request looks like a model that cannot help with it. Research on prefilled replies, a single internal refusal switch, and cheap fine-tuning suggests that safety training mostly shapes the first few words of a reply and leaves the underlying ability in place, and that telling a refusal from an inability is the harder problem.",{"src":739},{"authors":1002,"badge":1005,"source":1007},[1003],{"avatar":1004,"name":708,"to":709},{"src":707},{"label":1006},"AI Safety",{"name":713,"url":714},{"title":298,"description":999},"OwDhgJTHLO5IB5dU6_0UCNNGquc-PX8A4D-8pri9pbY",{"id":1011,"title":326,"body":1012,"date":1445,"description":1446,"extension":701,"image":1447,"meta":1449,"navigation":715,"path":327,"seo":1456,"stem":328,"__hash__":1457,"_path":327},"insights\u002Fnews\u002Finsights\u002Fthe-part-you-meant-to-keep.md",{"type":360,"value":1013,"toc":1435},[1014,1026,1029,1037,1041,1047,1050,1058,1062,1070,1078,1082,1090,1098,1101,1197,1204,1210,1214,1222,1230,1238,1244,1248,1251,1323,1326,1330,1336,1344,1347],[363,1015,367,1017,367,1021],{"className":1016},[366],[369,1018,326],{"className":1019,"id":1020},[372],"the-part-you-meant-to-keep",[376,1022,1025],{"className":1023,"id":1024},[379],"is-there-a-point-where-ai-help-turns-into-harm-and-could-a-scale-find-it","Is There a Point Where AI Help Turns Into Harm, and Could a Scale Find It?",[393,1027,1028],{},"Few people ask whether a letter went through spell-check. The small bottle of correction fluid that once sat beside the typewriter went away with the carbon paper, and the red underline that replaced it stopped being a topic of conversation long ago. A cover letter written with a chatbot still draws suspicion, and detectors and watermarks exist to answer whether a model was involved. As models settle into ordinary revision, a different question may matter more. Did the help cost anything, and at what point?",[393,1030,1031,1032,1036],{},"Psychologists call this cognitive offloading, using a physical action to make a task less demanding on the mind, like setting a phone reminder instead of memorizing a date ",[397,1033,1034],{},[400,1035,403],{"href":402},". Usually that is a fair trade. So where does a convenience start to become a loss? Is there a line, and could anyone measure it?",[376,1038,1040],{"id":1039},"the-cost-arrives-later","The Cost Arrives Later",[363,1042,385,1043],{"style":384},[387,1044],{"src":1045,"alt":1046,"style":391},"https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1793393444608-3886cf95b354?w=1200&auto=format&fit=crop","A lifter on a bench press with a spotter standing behind the bar",[393,1048,1049],{},"The spotter's hands rest at the hips until the bar stalls. A nudge on the last repetition finishes the set, and the lifter's chest still did the work. Lifting from the first repetition would move the same weight along the same path, and nothing about the bar would show the difference. AI help has a similar measurement problem. The output can look fine either way, and any cost tends to appear once the hands come away.",[393,1051,1052,1053,1057],{},"How quickly can that happen? Across three randomized experiments, people practiced fractions or reading comprehension in sessions of about a quarter of an hour, some with a chatbot beside them, then took a short test alone. The chatbot helped during practice. On the test, those in the first experiment who had used it solved 57 percent of problems, compared with 74 percent for those who had not, and they gave up more often ",[397,1054,1055],{},[400,1056,413],{"href":412},". The gap was smaller in a replication, and testing came right after practice, so the authors leave open whether the effect builds or fades with daily use.",[376,1059,1061],{"id":1060},"counting-requests-misses-it","Counting Requests Misses It",[393,1063,1064,1065,1069],{},"The obvious dial is quantity. Could some number of requests per task, or hours per week, mark the point of trouble? One small experiment points elsewhere. In it, 124 adults worked through logic puzzles, most with a simulated assistant whose price per request varied. Once starting ability was accounted for, request counts said almost nothing about skill gained. What tracked it was solo share, the fraction of time a person worked alone on a problem before asking ",[397,1066,1067],{},[400,1068,430],{"href":429},". Scores while assisted also overstated how people later did on their own, which suggests that a session going well is a poor guide.",[393,1071,1072,1073,1077],{},"A larger experiment on fraction practice gives the nearest thing to a dose curve. Each extra 10 percentage points of problems on which someone requested the full answer came with 32 percent lower odds of getting a later test question right ",[397,1074,1075],{},[400,1076,522],{"href":521},". That analysis is correlational, and it assumes a smooth trend, so a sharper break could be hiding inside it. Access alone produced no detectable harm in that study, perhaps because the assistant gave full answers only when explicitly asked. No study found in this research set out to look for a break point. For now a ramp seems at least as plausible as a cliff, which is inconvenient for anyone hoping to draw one line.",[376,1079,1081],{"id":1080},"five-rungs-drawn-in-pencil","Five Rungs, Drawn in Pencil",[393,1083,1084,1085,1089],{},"If quantity is a weak guide, depth looks stronger. In the replication of those short experiments, people who said they mostly asked for hints scored about the same as people with no chatbot, while those who mostly asked for answers fell behind ",[397,1086,1087],{},[400,1088,413],{"href":412},". Those groups chose their own habits, so the comparison only suggests a cause.",[393,1091,1092,1093,1097],{},"One recent writing study offers a vocabulary for depth. It separates four layers, surface (grammar and local wording), structural (outlines and organization), idea (claims and examples) and reasoning (evidence interpretation and argumentative logic), with coders rating each from 1, where the writer did the work, to 5, where the tool did it and little changed. Over an eight-week course, heavier offloading at every layer went with weaker thinking on later no-AI tasks, and the cost of open access ran most faintly through the surface layer and most strongly through reasoning ",[397,1094,1095],{},[400,1096,542],{"href":541},". Open-access writers also sent nearly twice as many prompts per task, yet far fewer of their revisions were writer-led, so more turns may not mean more involvement. Six classes without random assignment can only suggest so much, and the author says the framework still needs independent validation.",[393,1099,1100],{},"Adding memory at the bottom gives a rough ladder. It was assembled for this article, and no validation stands behind it.",[1102,1103,1104,1123],"table",{},[1105,1106,1107],"thead",{},[1108,1109,1110,1114,1117,1120],"tr",{},[1111,1112,1113],"th",{},"Rung",[1111,1115,1116],{},"What moves to the tool",[1111,1118,1119],{},"Usually helps when",[1111,1121,1122],{},"May start to cost when",[1124,1125,1126,1141,1155,1169,1183],"tbody",{},[1108,1127,1128,1132,1135,1138],{},[1129,1130,1131],"td",{},"1. Storing",[1129,1133,1134],{},"Dates, lists, references",[1129,1136,1137],{},"The facts can be looked up again",[1129,1139,1140],{},"Recall is the skill being built",[1108,1142,1143,1146,1149,1152],{},[1129,1144,1145],{},"2. Polishing",[1129,1147,1148],{},"Spelling, grammar, local wording",[1129,1150,1151],{},"The meaning is already settled",[1129,1153,1154],{},"Checking stops, or edits drift into meaning",[1108,1156,1157,1160,1163,1166],{},[1129,1158,1159],{},"3. Arranging",[1129,1161,1162],{},"Outline and order",[1129,1164,1165],{},"The person knows the material",[1129,1167,1168],{},"The person is still learning how the pieces fit",[1108,1170,1171,1174,1177,1180],{},[1129,1172,1173],{},"4. Supplying",[1129,1175,1176],{},"Claims, examples, solutions",[1129,1178,1179],{},"The task is routine",[1129,1181,1182],{},"Working it out was the point",[1108,1184,1185,1188,1191,1194],{},[1129,1186,1187],{},"5. Deciding",[1129,1189,1190],{},"Interpretation and judgment",[1129,1192,1193],{},"Stakes are low and the call is easy to check",[1129,1195,1196],{},"The person will answer for the conclusion later",[393,1198,1199,1200,431],{},"Even the second rung has a price. In a 2005 experiment, people with stronger verbal skills left nearly twice as many unflagged errors behind when a spelling and grammar checker was switched on ",[397,1201,1202],{},[400,1203,549],{"href":548},[393,1205,1206,1207,1209],{},"Read across the rows and the threshold stops being a place on the ladder. One working definition might be that help turns to harm at the rung where the tool starts doing something the person wanted, or needed, to still be able to do alone. A status email can sit at rung five with little lost. A junior engineer at rung four on debugging may be giving away what the job was supposed to build, the worry behind ",[400,1208,78],{"href":79},". Same ladder, opposite verdicts, so a useful scale would need to ask about intent as well as behavior.",[376,1211,1213],{"id":1212},"the-closest-thing-to-a-number","The Closest Thing to a Number",[393,1215,1216,1217,1221],{},"Could the rungs be measured automatically? The most direct attempt found here is the Offloading Score. It reads screen recordings and keystrokes, has a language model rebuild how an average person would finish each AI-assisted step without the tool, and reports the fraction of steps saved, between 0 and 1. Among 40 freelance developers, a one-hour limit produced more offloading than a four-hour limit, 0.45 against 0.31, while self-reported mental load barely moved ",[397,1218,1219],{},[400,1220,575],{"href":574},". People may not feel the shift as it happens.",[393,1223,1224,1225,1229],{},"The authors also sketch an actual line, an overreliance threshold of 0.33 tied to how well developers could recall their own system. They call it illustrative and say real thresholds should be set empirically. The link to recall was weak overall and stronger only after setting aside about 30 percent of users who combined moderate to high reliance with high recall, apparently using the tool to learn ",[397,1226,1227],{},[400,1228,575],{"href":574},". Those users could cross that line with no visible harm, and they may be the most interesting people in the study.",[393,1231,1232,1233,1237],{},"Questionnaires take the other route. A free self-check called PAUSE sorts readings into minimal, mild, moderate and strong bands, and its author is candid that the cut points were chosen for legibility, with no reliability or validity evidence behind them ",[397,1234,1235],{},[400,1236,873],{"href":872},". Levels are easy to print. Knowing what they mean is the hard part.",[393,1239,1240,1241,1243],{},"From a machine learning standpoint, this resembles choosing an operating point on a classifier. A score can come from logs, but the cut depends on what each kind of mistake costs, and those costs differ between an expense note and a thesis. The ground truth, what someone can do later with the tool gone, is slow and expensive to collect, and several studies above tested within minutes of practice. Then comes the feedback loop. Once anyone is judged on the number, prompts will likely get padded, the pattern ",[400,1242,270],{"href":271}," traced in systems that learn to satisfy their own metrics.",[376,1245,1247],{"id":1246},"right-use-wrong-use","Right Use, Wrong Use",[393,1249,1250],{},"Until a validated scale exists, the ladder can still sort everyday choices. Right use leaves a person able to do what they wanted to keep. Wrong use gives that ability away, often with little sign at the time. The labels depend on that goal, so the same request can land in either column. These pairings follow from the evidence above, and no study has tested them as a set.",[1102,1252,1253,1266],{},[1105,1254,1255],{},[1108,1256,1257,1260,1263],{},[1111,1258,1259],{},"Task",[1111,1261,1262],{},"Right use",[1111,1264,1265],{},"Wrong use",[1124,1267,1268,1279,1290,1301,1312],{},[1108,1269,1270,1273,1276],{},[1129,1271,1272],{},"Grammar and wording",[1129,1274,1275],{},"Polishing a draft with settled meaning, then checking the edits",[1129,1277,1278],{},"Accepting every change unread, even when the meaning shifts",[1108,1280,1281,1284,1287],{},[1129,1282,1283],{},"Learning something new",[1129,1285,1286],{},"Trying first, then asking for a hint",[1129,1288,1289],{},"Asking for the full answer before trying",[1108,1291,1292,1295,1298],{},[1129,1293,1294],{},"Organizing work",[1129,1296,1297],{},"Outlining familiar material",[1129,1299,1300],{},"Letting the tool arrange material still being learned",[1108,1302,1303,1306,1309],{},[1129,1304,1305],{},"Ideas and arguments",[1129,1307,1308],{},"Getting help on routine work that can be checked",[1129,1310,1311],{},"Submitting an argument the writer could not explain",[1108,1313,1314,1317,1320],{},[1129,1315,1316],{},"Decisions",[1129,1318,1319],{},"Handing off low-stakes calls that are easy to check",[1129,1321,1322],{},"Handing off a conclusion the person must answer for later",[393,1324,1325],{},"A rough test fits in one question. After the session, could the person explain, defend or reproduce the parts of the work they meant to keep? A yes suggests the tool supported them. A no may be fine for a status email, as long as the trade was made on purpose.",[376,1327,1329],{"id":1328},"a-bubble-between-two-lines","A Bubble Between Two Lines",[363,1331,385,1332],{"style":384},[387,1333],{"src":1334,"alt":1335,"style":391},"https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1786341594766-22826f989d0b?w=1200&auto=format&fit=crop","A yellow spirit level vial with its bubble between two black lines",[393,1337,1338,1339,1343],{},"A bubble between two lines straightens nothing. It tells the person holding the level where things stand, and the rest is up to them. That may be the more realistic job for a scale, and there is some early support for it. In the larger fraction experiment, some participants read a short note between problems describing what they had just done and what it meant for the practice they got, with no grades, points or instructions. Their odds of requesting a full answer fell by about half and they did better on the unaided test, while a points reward for effort changed neither ",[397,1340,1341],{},[400,1342,522],{"href":521},". The margins of error were wide, and it was one brief session with an immediate test, so durability remains unknown. It also hints at something awkward for anyone building a scale. The note seems to have worked without a score.",[393,1345,1346],{},"Spell-check seems to have stopped being a question once people had decided, mostly without discussing it, what they no longer needed to do themselves. That decision has not been made for reasoning, and it may not get made the same way for everyone. A scale might never find one point where help becomes harm. Could it still show a person where they stood on a given afternoon, close enough to change what they hand over next time?",[363,1348,367,1350,367,1352],{"className":1349},[580,581],[376,1351,584],{"id":580},[586,1353,385,1355,385,1366,385,1375,385,1386,385,1395,385,1406,385,1417,385,1426,367],{"className":1354},[589,590,591,592],[594,1356,1357,1358,1361,1362],{"id":596},"E. F. Risko and S. J. Gilbert, \"Cognitive Offloading,\" ",[599,1359,1360],{},"Trends in Cognitive Sciences",", vol. 20, no. 9, pp. 676–688, 2016. DOI: ",[400,1363,610],{"href":1364,"target":606,"className":1365},"https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.tics.2016.07.002",[608,609],[594,1367,1368,1369,921,1371],{"id":613},"G. Liu et al., \"AI Assistance Reduces Persistence and Hurts Independent Performance,\" ",[599,1370,920],{},[400,1372,610],{"href":1373,"target":606,"className":1374},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.04721",[608,609],[594,1376,1377,1378,1381,1382],{"id":625},"S. Wu et al., \"How AI Assistance Affects Human Skill Development: A Study of Learning with Logic Puzzles,\" in ",[599,1379,1380],{},"Proc. ACM Conference on Human-AI Complementarity and Alignment (HCOMP 2026)",", 2026. DOI: ",[400,1383,610],{"href":1384,"target":606,"className":1385},"https:\u002F\u002Fdoi.org\u002F10.1145\u002F3834580.3838741",[608,609],[594,1387,1388,1389,921,1391],{"id":637},"S. Maier et al., \"Designing Against Deskilling: Metacognitive Feedback Reduces Cognitive Offloading to LLM Assistants,\" ",[599,1390,920],{},[400,1392,610],{"href":1393,"target":606,"className":1394},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.20143",[608,609],[594,1396,1397,1398,1401,1402],{"id":649},"X. Chen, \"Layer-sensitive cognitive offloading in generative AI-assisted writing: supported performance and independent no-AI outcomes,\" ",[599,1399,1400],{},"Frontiers in Psychology",", vol. 17, art. 1906199, 2026. DOI: ",[400,1403,610],{"href":1404,"target":606,"className":1405},"https:\u002F\u002Fdoi.org\u002F10.3389\u002Ffpsyg.2026.1906199",[608,609],[594,1407,1408,1409,1412,1413],{"id":661},"D. F. Galletta et al., \"Does Spell-Checking Software Need a Warning Label?,\" ",[599,1410,1411],{},"Communications of the ACM",", vol. 48, no. 7, pp. 82–86, 2005. DOI: ",[400,1414,610],{"href":1415,"target":606,"className":1416},"https:\u002F\u002Fdoi.org\u002F10.1145\u002F1070838.1070841",[608,609],[594,1418,1419,1420,921,1422],{"id":673},"V. Padmakumar et al., \"Offloading Score: Measuring AI Reliance Through Counterfactual Workflows,\" ",[599,1421,920],{},[400,1423,610],{"href":1424,"target":606,"className":1425},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.29392",[608,609],[594,1427,1428,1429,921,1431],{"id":979},"M. U. Alam, \"PAUSE: A Privacy-Preserving Self-Reflection Tool for AI-Associated Cognitive Offloading,\" ",[599,1430,920],{},[400,1432,610],{"href":1433,"target":606,"className":1434},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.13155",[608,609],{"title":684,"searchDepth":685,"depth":685,"links":1436},[1437,1438,1439,1440,1441,1442,1443,1444],{"id":1024,"depth":685,"text":1025},{"id":1039,"depth":685,"text":1040},{"id":1060,"depth":685,"text":1061},{"id":1080,"depth":685,"text":1081},{"id":1212,"depth":685,"text":1213},{"id":1246,"depth":685,"text":1247},{"id":1328,"depth":685,"text":1329},{"id":580,"depth":685,"text":584},"2026-11-02","AI help tends to look good in the moment and cost something later. The evidence so far offers no single cutoff, and the line seems to move with what a person wants to keep. Could a cognitive offloading scale find it?",{"src":1448},"https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1530686350401-7de25243dd89?w=1200&auto=format&fit=crop",{"authors":1450,"badge":1453,"source":1455},[1451],{"avatar":1452,"name":708,"to":709},{"src":707},{"label":1454},"Cognitive Offloading",{"name":713,"url":714},{"title":326,"description":1446},"I2CiH4E-x_25GqPZ7G3zu0bgjJ6xoQJ8ikXT2zhnVQ4",1794600584666]