--- title: “We were looking at it wrong”: The explosive claim... description: Nick Cammarata, a researcher at AI startup Goodfire, published a 26,000-word blog post claiming neural networks may be far more interpretable than previously... canonical: https://daily.dev/posts/we-were-looking-at-it-wrong-the-explosive-claim-that-s-upended-the-ai-world-sx0b40sqo twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: “We were looking at it wrong”: The explosive claim that’s upended the AI world | daily.dev og:description: Nick Cammarata, a researcher at AI startup Goodfire, published a 26,000-word blog post claiming neural networks may be far more interpretable than previously... og:url: https://daily.dev/posts/we-were-looking-at-it-wrong-the-explosive-claim-that-s-upended-the-ai-world-sx0b40sqo og:image: https://api.daily.dev/og/posts/sX0B40SQO.png og:image:alt: “We were looking at it wrong”: The explosive claim that’s upended the AI world 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. # “We were looking at it wrong”: The explosive claim that’s upended the AI world **[MIT Technology Review](https://daily.dev/sources/technologyreview)** · 10 min read · 0 upvotes · 0 comments ## Summary Nick Cammarata, a researcher at AI startup Goodfire, published a 26,000-word blog post claiming neural networks may be far more interpretable than previously believed, sparking strong reactions across the AI research community. His key insight came from revisiting a 2020 finding about a neuron in the image model Inception v1 that responded to both cars and cats—he realized both share a 'V' shape, suggesting networks encode legible patterns rather than arbitrary tangled superposition. Cammarata now believes full understanding of LLMs may eventually be achievable with enough effort, which would have major implications for AI safety and control. However, prominent researchers including Chris Potts, Jacob Hilton, and Neel Nanda pushed back, arguing the claims are overhyped, noting current understanding of top LLMs covers less than 0.01% of their inner workings, and that neuron-level insight doesn't address how neurons combine into circuits. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What did Nick Cammarata discover about the car-and-cat neuron in Inception v1? The neuron responded to both cars and cats because both often feature a prominent V-shaped pattern, such as a cat's face or the front of a car viewed from the side. Cammarata also found the neuron activated for a pool table corner, a train's pointed nose, and a fire boat spraying water, while round O-shaped images turned it off. _Developers tracking interpretability breakthroughs can follow debates like this one on daily.dev._ ### How much of a top-tier LLM's inner workings do researchers currently understand through mechanistic interpretability? Researchers understand less than 0.01% of the inner workings of top-tier large language models, according to Jacob Hilton of the Alignment Research Center. He argues that even promising neuron-level findings don't yet explain how neurons connect into larger circuits, so claims of a major breakthrough in full model understanding are premature. _Anyone weighing AI safety claims against actual research progress can track this discussion on daily.dev._ ### What is the superposition hypothesis in neural network interpretability research? Superposition is the idea that neural networks pack multiple unrelated concepts into single neurons, discovered when researchers studying Inception v1 in 2020 found a neuron responding to both cars and cats. This phenomenon made models appear as tangled, incomprehensible knots and became the basis for most interpretability research focused on untangling that complexity, though Nick Cammarata now argues the packed concepts actually share hidden patterns. _Engineers curious about why models resist easy interpretation can follow this debate on daily.dev._ ## Similar posts on daily.dev - [OpenAI’s new LLM exposes the secrets of how AI really works](https://daily.dev/posts/openai-s-new-llm-exposes-the-secrets-of-how-ai-really-works-iepupajzg) · MIT Technology Review · 0 upvotes · 0 comments - [Can We Understand How Large Language Models Reason? – Communications of the ACM](https://daily.dev/posts/can-we-understand-how-large-language-models-reason-communications-of-the-acm-pcrymfnhd) · Hacker News · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#neural-networks](https://daily.dev/tags/neural-networks), [#ai-safety](https://daily.dev/tags/ai-safety) [View this post on daily.dev](https://daily.dev/posts/we-were-looking-at-it-wrong-the-explosive-claim-that-s-upended-the-ai-world-sx0b40sqo) ```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":"“We were looking at it wrong”: The explosive claim that’s upended the AI world","url":"https://daily.dev/posts/we-were-looking-at-it-wrong-the-explosive-claim-that-s-upended-the-ai-world-sx0b40sqo","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/we-were-looking-at-it-wrong-the-explosive-claim-that-s-upended-the-ai-world-sx0b40sqo"},"datePublished":"2026-10-19T18:08:39.999Z","dateModified":"2026-10-19T18:09:44.641Z","description":"Nick Cammarata, a researcher at AI startup Goodfire, published a 26,000-word blog post claiming neural networks may be far more interpretable than previously...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/2ba6298c45e8673319931a6fc12a374e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/2ba6298c45e8673319931a6fc12a374e?_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":0,"discussionUrl":"https://daily.dev/posts/we-were-looking-at-it-wrong-the-explosive-claim-that-s-upended-the-ai-world-sx0b40sqo","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,neural-networks,ai-safety","timeRequired":"PT10M"} {"@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":"“We were looking at it wrong”: The explosive claim that’s upended the AI world"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/we-were-looking-at-it-wrong-the-explosive-claim-that-s-upended-the-ai-world-sx0b40sqo#faq","mainEntity":[{"@type":"Question","name":"What did Nick Cammarata discover about the car-and-cat neuron in Inception v1?","acceptedAnswer":{"@type":"Answer","text":"The neuron responded to both cars and cats because both often feature a prominent V-shaped pattern, such as a cat's face or the front of a car viewed from the side. Cammarata also found the neuron activated for a pool table corner, a train's pointed nose, and a fire boat spraying water, while round O-shaped images turned it off. Developers tracking interpretability breakthroughs can follow debates like this one on daily.dev."}},{"@type":"Question","name":"How much of a top-tier LLM's inner workings do researchers currently understand through mechanistic interpretability?","acceptedAnswer":{"@type":"Answer","text":"Researchers understand less than 0.01% of the inner workings of top-tier large language models, according to Jacob Hilton of the Alignment Research Center. He argues that even promising neuron-level findings don't yet explain how neurons connect into larger circuits, so claims of a major breakthrough in full model understanding are premature. Anyone weighing AI safety claims against actual research progress can track this discussion on daily.dev."}},{"@type":"Question","name":"What is the superposition hypothesis in neural network interpretability research?","acceptedAnswer":{"@type":"Answer","text":"Superposition is the idea that neural networks pack multiple unrelated concepts into single neurons, discovered when researchers studying Inception v1 in 2020 found a neuron responding to both cars and cats. This phenomenon made models appear as tangled, incomprehensible knots and became the basis for most interpretability research focused on untangling that complexity, though Nick Cammarata now argues the packed concepts actually share hidden patterns. Engineers curious about why models resist easy interpretation can follow this debate on daily.dev."}}]} ```