--- title: Don’t be fooled—LLMs don’t reason | daily.dev description: A former DeepMind AlphaGo researcher argues that today's large language models, even with chain-of-thought prompting, don't actually reason the way AlphaGo did... canonical: https://daily.dev/posts/don-t-be-fooled-llms-don-t-reason-4waw8d0j4 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Don’t be fooled—LLMs don’t reason | daily.dev og:description: A former DeepMind AlphaGo researcher argues that today's large language models, even with chain-of-thought prompting, don't actually reason the way AlphaGo did... og:url: https://daily.dev/posts/don-t-be-fooled-llms-don-t-reason-4waw8d0j4 og:image: https://api.daily.dev/og/posts/4WAW8d0j4.png og:image:alt: Don’t be fooled—LLMs don’t reason og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Don’t be fooled—LLMs don’t reason **[MIT Technology Review](https://daily.dev/sources/technologyreview)** · 8 min read · 0 upvotes · 1 comments ## Summary A former DeepMind AlphaGo researcher argues that today's large language models, even with chain-of-thought prompting, don't actually reason the way AlphaGo did against Lee Sedol in 2016. AlphaGo combined a fast 'intuitive' policy network with an explicit, inspectable search over a game tree of future possibilities—a genuine system 1/system 2 split. LLMs, by contrast, only generate longer sequences of next-token predictions; their chain-of-thought output lacks a persistent, auditable epistemic state, conflates knowledge with reasoning inside the model weights, and research shows models often fabricate a plausible-sounding reasoning trace after already reaching an answer. The author, who recently left DeepMind, proposes building systems with an explicit epistemic state—tracking settled facts, doubts, and open questions—updated only when backed by evidence, as a path toward trustworthy reasoning for high-stakes domains like medicine and science. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why isn't chain-of-thought in LLMs considered real reasoning? Chain-of-thought output is still produced by the same next-token prediction process rather than a separate reasoning mechanism like AlphaGo's game-tree search. It lacks a persistent, inspectable epistemic state tracking hypotheses and evidence, conflates knowledge with reasoning inside the model's weights, and research has shown models often construct the reasoning trace after already reaching an answer rather than genuinely following it. _Developers evaluating LLM reliability for complex tasks can follow this reasoning debate on daily.dev._ ### How did AlphaGo's move 37 against Lee Sedol actually get chosen? AlphaGo's policy network rated move 37 as having only a roughly one-in-10,000 chance of being played by an expert, but its search machinery explicitly built and evaluated a game tree with thousands of branches representing possible futures, and that deliberate search is what surfaced the move as superior despite looking implausible at a glance. _Anyone curious how search-based AI differs from LLM pattern prediction can track this topic on daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@whlsk34** · 0 upvotes > Hot take: what if they pseudo-reason and that's good enough to do all sorts of crazy things. Maybe it's my fault for reading articles that are meant for the layperson, but every time I see one of these 'LLMs dont/don't have X' I really hope and then pine for someone who defines their terms well, gives lots of intuitive examples in cognitive science and explains the phenomenology of LLMs. Still looking for that article. ## Similar posts on daily.dev - [Water Cooler Small Talk, Ep. 9: What “Thinking” and “Reasoning” Really Mean in AI and LLMs](https://daily.dev/posts/water-cooler-small-talk-ep-9-what-thinking-and-reasoning-really-mean-in-ai-and-llms-dqzo11qq1) · Towards Data Science · 0 upvotes · 0 comments - [Google DeepMind wants to know if chatbots are just virtue signaling](https://daily.dev/posts/google-deepmind-wants-to-know-if-chatbots-are-just-virtue-signaling-y0n7mzotz) · MIT Technology Review · 1 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#prompt-engineering](https://daily.dev/tags/prompt-engineering) [View this post on daily.dev](https://daily.dev/posts/don-t-be-fooled-llms-don-t-reason-4waw8d0j4) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Don’t be fooled—LLMs don’t reason","url":"https://daily.dev/posts/don-t-be-fooled-llms-don-t-reason-4waw8d0j4","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/don-t-be-fooled-llms-don-t-reason-4waw8d0j4"},"datePublished":"2026-10-02T08:34:49.866Z","dateModified":"2026-10-02T08:35:14.966Z","description":"A former DeepMind AlphaGo researcher argues that today's large language models, even with chain-of-thought prompting, don't actually reason the way AlphaGo did...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9276d186464fe8a55f7b04c2deafd0b9?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9276d186464fe8a55f7b04c2deafd0b9?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"MIT Technology Review","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"MIT Technology Review","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/a8d03889722f4f22b3986e0b395ff076","url":"https://daily.dev/sources/technologyreview"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/don-t-be-fooled-llms-don-t-reason-4waw8d0j4","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"ai,prompt-engineering","timeRequired":"PT8M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"MIT Technology Review","item":"https://daily.dev/sources/technologyreview"},{"@type":"ListItem","position":3,"name":"Don’t be fooled—LLMs don’t reason"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/don-t-be-fooled-llms-don-t-reason-4waw8d0j4","comment":[{"@type":"Comment","text":"Hot take: what if they pseudo-reason and that’s good enough to do all sorts of crazy things. Maybe it’s my fault for reading articles that are meant for the layperson, but every time I see one of these ‘LLMs dont/don’t have X’ I really hope and then pine for someone who defines their terms well, gives lots of intuitive examples in cognitive science and explains the phenomenology of LLMs. Still looking for that article.","datePublished":"2026-10-03T08:21:39.650Z","url":"https://daily.dev/posts/4WAW8d0j4#c-NVRhX0WKc","author":{"@type":"Person","name":"whlsk34","url":"https://daily.dev/whlsk34","image":"https://media.daily.dev/image/upload/s--O0TOmw4y--/f_auto/v1715772965/public/noProfile"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/don-t-be-fooled-llms-don-t-reason-4waw8d0j4#faq","mainEntity":[{"@type":"Question","name":"Why isn't chain-of-thought in LLMs considered real reasoning?","acceptedAnswer":{"@type":"Answer","text":"Chain-of-thought output is still produced by the same next-token prediction process rather than a separate reasoning mechanism like AlphaGo's game-tree search. It lacks a persistent, inspectable epistemic state tracking hypotheses and evidence, conflates knowledge with reasoning inside the model's weights, and research has shown models often construct the reasoning trace after already reaching an answer rather than genuinely following it. Developers evaluating LLM reliability for complex tasks can follow this reasoning debate on daily.dev."}},{"@type":"Question","name":"How did AlphaGo's move 37 against Lee Sedol actually get chosen?","acceptedAnswer":{"@type":"Answer","text":"AlphaGo's policy network rated move 37 as having only a roughly one-in-10,000 chance of being played by an expert, but its search machinery explicitly built and evaluated a game tree with thousands of branches representing possible futures, and that deliberate search is what surfaced the move as superior despite looking implausible at a glance. Anyone curious how search-based AI differs from LLM pattern prediction can track this topic on daily.dev."}}]} ```