[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-study-finds-llms-process-repeated-words-differently-after-training":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},5310,"study-finds-llms-process-repeated-words-differently-after-training","Study Finds LLMs Process Repeated Words Differently After Training","A study using repetition priming finds base LLMs reuse old representations automatically, while instruction-tuned models recompute them deliberately.","Language models handle repeated words very differently depending on whether they have been instruction-tuned, a new study finds.\n\nResearchers tested 15 models across five model families, ranging from 1.5 billion to 14 billion parameters, using a decades-old psychology technique called repetition priming, alongside human participants given identical material. They ran everything through two tasks: semantic categorization and cloze completion. Base models - the raw, pre-instruction versions - showed automatic processing: a speed boost on repeated words that held steady no matter the gap between repetitions, and that partly survived even when surrounding context was stripped out. Instruction-tuned models behaved differently: their speed boost faded as the gap grew, vanished without the expected context, and at larger scales flipped into outright interference, meaning repetition slowed them down instead of speeding them up.\n\nThat matters because instruction-tuning is the step nearly every deployed chatbot goes through, and this suggests it does more than teach politeness - it appears to change how the model retrieves and reuses information at a mechanical level. Within the Qwen 2.5 family, the gap between base and instruct behavior grew stronger with model size, which hints that scaling up instruction-tuned models is not a neutral operation.\n\nHumans, tested on the same stimuli, landed in between: sensitive to timing like the instruct models, but without the slowdown. Neither type of model fully explains how our own brains handle a familiar word showing up twice.","[\"llms\",\"ai-research\",\"cognitive-science\",\"nlp\"]","2026-08-18T04:00:00.000Z","2026-08-18T14:33:22.644Z","2026-08-18T14:33:34.422Z","published",null,[],"ai",[26,27,28,29],"llms","ai-research","cognitive-science","nlp",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.14681",0,{"sections":36},[37,41,45,50,55,60,65,70,75,79,84,89,94,99],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",3293,"2026-08-20T04:00:00.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":40},"Security","security",435,{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",210,"2026-08-19T09:32:27.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",140,"2026-08-19T18:25:42.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",95,"2026-08-18T16:05:00.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Science","science",90,"2026-08-19T18:41:02.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Software","software",73,"2026-08-18T07:51:50.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":18},"Dev Tools","dev-tools",69,{"name":80,"slug":81,"count":82,"latest_published_at":83},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":85,"slug":86,"count":87,"latest_published_at":88},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",33,"2026-08-18T22:18:13.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]