[{"data":1,"prerenderedAt":197},["ShallowReactive",2],{"learn-concept-/learn/context-language-models":3},{"id":4,"title":5,"body":6,"cardImage":158,"cardImageAlt":159,"date":160,"description":161,"domain":162,"domainKey":163,"extension":164,"featured":165,"fullName":5,"interaction":166,"maturity":167,"mentalModel":168,"meta":169,"navigation":170,"neighbors":171,"ogImage":166,"path":180,"published":170,"robots":166,"seo":181,"shortName":182,"sitemap":183,"socialImage":184,"socialImageAlt":159,"sources":185,"stem":190,"tags":191,"translationKey":195,"updated":160,"__hash__":196},"learnEn/learn/context-language-models.md","Context Language Models",{"type":7,"value":8,"toc":151},"minimal",[9,18,21,26,29,40,44,47,113,120,129,133,136,142,145],[10,11,12,13,17],"p",{},"Imagine a father, mother, and daughter asking an AI to plan three days in Tokyo. Their ",[14,15,16],"strong",{},"total budget is US$3,000",", and their daughter tires on long walks. Then they ask: “Make it five days.”",[10,19,20],{},"The duration changed. The budget and walking limit still apply. The next answer needs to account for all of them.",[22,23,25],"h2",{"id":24},"context-is-the-material-supplied-to-this-call","Context is the material supplied to this call",[10,27,28],{},"A model generates an answer from the instructions, conversation, and tool results supplied to that call. Think of those as papers on a desk: there may be many old plans, but the current job needs to be clear.",[10,30,31,32,39],{},"In September 2026, Rulin Shao and colleagues introduced ",[33,34,38],"a",{"href":35,"rel":36},"https://arxiv.org/abs/2609.37725",[37],"nofollow","Context Language Models (CLM)",", which let a model edit its own working context and organize the material carried into later work.",[22,41,43],{"id":42},"update-the-days-keep-the-other-conditions","Update the days. Keep the other conditions.",[10,45,46],{},"For this hypothetical trip, an updated working record could look like this:",[48,49,50,66],"table",{},[51,52,53],"thead",{},[54,55,56,60,63],"tr",{},[57,58,59],"th",{},"Condition",[57,61,62],{},"Before",[57,64,65],{},"Now",[67,68,69,83,93,103],"tbody",{},[54,70,71,75,78],{},[72,73,74],"td",{},"Duration",[72,76,77],{},"3 days",[72,79,80],{},[14,81,82],{},"5 days",[54,84,85,88,91],{},[72,86,87],{},"Total budget",[72,89,90],{},"US$3,000",[72,92,90],{},[54,94,95,98,101],{},[72,96,97],{},"Family",[72,99,100],{},"Dad, mom, daughter",[72,102,100],{},[54,104,105,108,111],{},[72,106,107],{},"Walking",[72,109,110],{},"Keep walks short",[72,112,110],{},[10,114,115,116,119],{},"It should also retain new unfinished work: ",[14,117,118],{},"recalculate accommodation and transport for two extra nights."," Updating a field does not prove the longer trip fits the budget.",[10,121,122,123,128],{},"The record must reach the next call. The official implementation mirrors editable context into a file, lets the model change it, and reads accepted changes back. System instructions and the original task stay pinned separately. The ",[33,124,127],{"href":125,"rel":126},"https://github.com/facebookresearch/context-language-models/blob/main/clm/clm_harness/README.md",[37],"official harness documentation"," explains this handoff.",[22,130,132],{"id":131},"a-habit-you-can-borrow","A habit you can borrow",[10,134,135],{},"When a task changes, ask:",[137,138,139],"blockquote",{},[10,140,141],{},"Before continuing, list what changed, which conditions still apply, and what new work the change creates. Then continue using the updated requirements.",[10,143,144],{},"This helps you inspect task state. A prompt alone does not activate CLM, and the model can still omit a condition. The trip is an illustration, not a measured result.",[10,146,147,148],{},"To check whether an AI has kept up, start with the material it will use next: ",[14,149,150],{},"does it still accurately describe the job you now want done?",{"title":152,"searchDepth":153,"depth":153,"links":154},"",2,[155,156,157],{"id":24,"depth":153,"text":25},{"id":42,"depth":153,"text":43},{"id":131,"depth":153,"text":132},"/learn-img/context-language-models/card-4x5.jpg","An illustrated Tokyo family trip record changes from three to five days, retains a US$3,000 budget and short-walk constraint, and feeds the updated record into the next AI call.","2026-10-06","Keep an editable working record so the next AI call receives the requirements that still apply.","AI systems","ai-systems","md",false,null,"growing","The trip gets longer. The budget and the child’s needs stay. The next call should see all three.",{},true,[172,176],{"name":173,"fullName":173,"category":174,"summary":175},"Context Window","input boundary","How much input a model can process in one call. A larger window still needs clear current requirements.",{"name":177,"fullName":177,"category":178,"summary":179},"Task State","working record","The current goal, active constraints, decisions already made, and unfinished work.","/learn/context-language-models",{"title":5,"description":161},"CLM",{"loc":180},"/learn-img/context-language-models/og-1200x627.jpg",[186,188],{"title":187,"url":35},"Rulin Shao et al. · Context Language Models",{"title":189,"url":125},"Official CLM harness · How a turn works","learn/context-language-models",[192,193,194],"context-management","AI-agents","working-context","context-language-models","P_bgamhO3LoIwLhVihC6hhY_rF2Y4vALuYBQ2xfgaUc",1792071062083]