{ "source_file": "0930", "source_platform": "Medium", "author": "Cristian Leo", "title_listed": "How LLMs Think", "keywords": "", "domain": "大模型算法", "content_type": "深度技术解析", "is_paid_flag": "否", "url": "https://towardsdatascience.com/how-llms-think-d8754a79017d/", "publish_date": "2024-06-07", "corpus_stage": "", "notes": "", "id": "6c6fc8bbc623", "collected_at": "2026-10-02T06:12:25", "title": "How LLMs Think | Towards Data Science", "body_markdown": "![Revisit consent button](../images/body-df29beacd7496220cc94.svg)\nWe value your privacy\nWe use cookies to enhance your browsing experience, serve personalised ads or content, and analyse our traffic. By clicking \"Accept All\", you consent to our use of cookies.\nCustomise Reject All Accept All\nCustomise Consent Preferences ![](../images/body-2159bb67f3de8ad870eb.svg)\nWe use cookies to help you navigate efficiently and perform certain functions. You will find detailed information about all cookies under each consent category below.\nThe cookies that are categorised as \"Necessary\" are stored on your browser as they are essential for enabling the basic functionalities of the site. ... Show more\nNecessaryAlways Active\nNecessary cookies are required to enable the basic features of this site, such as providing secure log-in or adjusting your consent preferences. These cookies do not store any personally identifiable data.\n * Cookie\nBCTempID\n * Duration\n10 minutes\n * Description\nNo description available.\n\n\n * Cookie\n__cf_bm\n * Duration\n1 hour\n * Description\nThis cookie, set by Cloudflare, is used to support Cloudflare Bot Management. \n\n\n * Cookie\nAWSALBCORS\n * Duration\n7 days\n * Description\nAmazon Web Services set this cookie for load balancing.\n\n\n * Cookie\n_cfuvid\n * Duration\nsession\n * Description\nCloudflare sets this cookie to track users across sessions to optimize user experience by maintaining session consistency and providing personalized services\n\n\n * Cookie\nli_gc\n * Duration\n6 months\n * Description\nLinkedin set this cookie for storing visitor's consent regarding using cookies for non-essential purposes.\n\n\n * Cookie\n__hssrc\n * Duration\nsession\n * Description\nThis cookie is set by Hubspot whenever it changes the session cookie. The __hssrc cookie set to 1 indicates that the user has restarted the browser, and if the cookie does not exist, it is assumed to be a new session.\n\n\n * Cookie\n__hssc\n * Duration\n1 hour\n * Description\nHubSpot sets this cookie to keep track of sessions and to determine if HubSpot should increment the session number and timestamps in the __hstc cookie.\n\n\n * Cookie\nwpEmojiSettingsSupports\n * Duration\nsession\n * Description\nWordPress sets this cookie when a user interacts with emojis on a WordPress site. It helps determine if the user's browser can display emojis properly.\n\n\n * Cookie\nBCSessionID\n * Duration\n1 year 1 month 4 days\n * Description\nBlueconic sets this cookie as a unique identifier for the BlueConic profile.\n\n\n * Cookie\n_octo\n * Duration\n1 year\n * Description\nNo description available.\n\n\n * Cookie\nlogged_in\n * Duration\n1 year\n * Description\nNo description available.\n\n\n * Cookie\n__Secure-YEC\n * Duration\npast\n * Description\nYouTube sets this cookie to stores the user's video player preferences using embedded YouTube video\n\n\n * Cookie\n__eoi\n * Duration\n6 months\n * Description\nDescription is currently not available.\n\n\n * Cookie\nAWSALBTGCORS\n * Duration\n7 days\n * Description\nNo description available.\n\n\n * Cookie\nlogin-status-p\n * Duration\npast\n * Description\nDescription is currently not available.\n\n\n * Cookie\nAWSALBTG\n * Duration\n7 days\n * Description\nNo description available.\n\n\n * Cookie\ncsrf_token\n * Duration\nsession\n * Description\nNo description available.\n\n\n * Cookie\ntoken_v2\n * Duration\n1 day\n * Description\nDescription is currently not available.\n\n\n * Cookie\nD\n * Duration\n1 year\n * Description\nDescription is currently not available.\n\n\n * Cookie\nPHPSESSID\n * Duration\nsession\n * Description\nThis cookie is native to PHP applications. The cookie stores and identifies a user's unique session ID to manage user sessions on the website. The cookie is a session cookie and will be deleted when all the browser windows are closed.\n\n\n * Cookie\nVISITOR_PRIVACY_METADATA\n * Duration\n6 months\n * Description\nYouTube sets this cookie to store the user's cookie consent state for the current domain. \n\n\n * Cookie\ncookietest\n * Duration\nsession\n * Description\nThe cookietest cookie is typically used to determine whether the user's browser accepts cookies, essential for website functionality and user experience.\n\n\n * Cookie\n__Host-airtable-session\n * Duration\n1 year\n * Description\nThis cookie is used to enable us to integrate the services of Airtable.\n\n\n * Cookie\n__Host-airtable-session.sig\n * Duration\n1 year\n * Description\nThis cookie is used to enable us to integrate the services of Airtable.\n\n\n * Cookie\nm\n * Duration\n1 year 1 month 4 days\n * Description\nStripe sets this cookie for fraud prevention purposes. It identifies the device used to access the website, allowing the website to be formatted accordingly.\n\n\n * Cookie\nBIGipServer*\n * Duration\nsession\n * Description\nMarketo sets this cookie to collect information about the user's online activity and build a profile about their interests to provide advertisements relevant to the user. \n\n\n * Cookie\n__cfruid\n * Duration\nsession\n * Description\nCloudflare sets this cookie to identify trusted web traffic.\n\n\n * Cookie\n_GRECAPTCHA\n * Duration\n6 months\n * Description\nGoogle Recaptcha service sets this cookie to identify bots to protect the website against malicious spam attacks.\n\n\n * Cookie\n__Secure-YNID\n * Duration\n6 months\n * Description\nGoogle cookie used to protect user security and prevent fraud, especially during the login process.\n\n\n * Cookie\ncookieyes-consent\n * Duration\n1 year\n * Description\nCookieYes sets this cookie to remember users' consent preferences so that their preferences are respected on subsequent visits to this site. It does not collect or store any personal information about the site visitors.\n\n\nFunctional\nFunctional cookies help perform certain functionalities like sharing the content of the website on social media platforms, collecting feedback, and other third-party features.\n * Cookie\nlidc\n * Duration\n1 day\n * Description\nLinkedIn sets the lidc cookie to facilitate data center selection.\n\n\n * Cookie\nbrw\n * Duration\n1 year\n * Description\nNo description available.\n\n\n * Cookie\nbrwConsent\n * Duration\n5 minutes\n * Description\nDescription is currently not available.\n\n\n * Cookie\nWMF-Uniq\n * Duration\n1 year\n * Description\nDescription is currently not available.\n\n\n * Cookie\nloom_anon_comment\n * Duration\n1 year\n * Description\nNo description available.\n\n\n * Cookie\nloom_referral_video\n * Duration\nsession\n * Description\nDescription is currently not available.\n\n\n * Cookie\nVISITOR_INFO1_LIVE\n * Duration\n6 months\n * Description\nA cookie set by YouTube to measure bandwidth that determines whether the user gets the new or old player interface.\n\n\n * Cookie\nyt-remote-connected-devices\n * Duration\nNever Expires\n * Description\nYouTube sets this cookie to store the user's video preferences using embedded YouTube videos.\n\n\n * Cookie\nytidb::LAST_RESULT_ENTRY_KEY\n * Duration\nNever Expires\n * Description\nThe cookie ytidb::LAST_RESULT_ENTRY_KEY is used by YouTube to store the last search result entry that was clicked by the user. This information is used to improve the user experience by providing more relevant search results in the future.\n\n\n * Cookie\nyt-remote-device-id\n * Duration\nNever Expires\n * Description\nYouTube sets this cookie to store the user's video preferences using embedded YouTube videos.\n\n\n * Cookie\nyt-remote-session-name\n * Duration\nsession\n * Description\nThe yt-remote-session-name cookie is used by YouTube to store the user's video player preferences using embedded YouTube video.\n\n\n * Cookie\nyt-remote-fast-check-period\n * Duration\nsession\n * Description\nThe yt-remote-fast-check-period cookie is used by YouTube to store the user's video player preferences for embedded YouTube videos.\n\n\n * Cookie\nyt-remote-session-app\n * Duration\nsession\n * Description\nThe yt-remote-session-app cookie is used by YouTube to store user preferences and information about the interface of the embedded YouTube video player.\n\n\n * Cookie\nyt-remote-cast-available\n * Duration\nsession\n * Description\nThe yt-remote-cast-available cookie is used to store the user's preferences regarding whether casting is available on their YouTube video player.\n\n\n * Cookie\nyt-remote-cast-installed\n * Duration\nsession\n * Description\nThe yt-remote-cast-installed cookie is used to store the user's video player preferences using embedded YouTube video.\n\n\n * Cookie\ncp_session\n * Duration\n3 months\n * Description\nCodepen sets this cookie for Help systems found in the website.\n\n\n * Cookie\nloid\n * Duration\n1 year 1 month 4 days\n * Description\nThis cookie is set by the Reddit. The cookie enables the sharing of content from the website onto the social media platform.\n\n\nAnalytics\nAnalytical cookies are used to understand how visitors interact with the website. These cookies help provide information on metrics such as the number of visitors, bounce rate, traffic source, etc.\n * Cookie\n__hstc\n * Duration\n6 months\n * Description\nHubspot set this main cookie for tracking visitors. It contains the domain, initial timestamp (first visit), last timestamp (last visit), current timestamp (this visit), and session number (increments for each subsequent session).\n\n\n * Cookie\nhubspotutk\n * Duration\n6 months\n * Description\nHubSpot sets this cookie to keep track of the visitors to the website. This cookie is passed to HubSpot on form submission and used when deduplicating contacts.\n\n\n * Cookie\n_ga\n * Duration\n1 year 1 month 4 days\n * Description\nGoogle Analytics sets this cookie to calculate visitor, session and campaign data and track site usage for the site's analytics report. The cookie stores information anonymously and assigns a randomly generated number to recognise unique visitors.\n\n\n * Cookie\n_ga_*\n * Duration\n1 year 1 month 4 days\n * Description\nGoogle Analytics sets this cookie to store and count page views.\n\n\n * Cookie\n__Host-psifi.analyticsTrace\n * Duration\n6 hours\n * Description\nDescription is currently not available.\n\n\n * Cookie\n__Host-psifi.analyticsTraceV2\n * Duration\n6 hours\n * Description\nDescription is currently not available.\n\n\n * Cookie\n_gh_sess\n * Duration\nsession\n * Description\nGitHub sets this cookie for temporary application and framework state between pages like what step the user is on in a multiple step form.\n\n\n * Cookie\nYSC\n * Duration\nsession\n * Description\nYSC cookie is set by Youtube and is used to track the views of embedded videos on Youtube pages.\n\n\n * Cookie\najs_anonymous_id\n * Duration\n1 year\n * Description\nThis cookie is set by Segment to count the number of people who visit a certain site by tracking if they have visited before.\n\n\n * Cookie\nvuid\n * Duration\n1 year 1 month 4 days\n * Description\nVimeo installs this cookie to collect tracking information by setting a unique ID to embed videos on the website. \n\n\nPerformance\nPerformance cookies are used to understand and analyse the key performance indexes of the website which helps in delivering a better user experience for the visitors.\n * Cookie\nAWSALB\n * Duration\n7 days\n * Description\nAWSALB is an application load balancer cookie set by Amazon Web Services to map the session to the target.\n\n\n * Cookie\nacq\n * Duration\npast\n * Description\nDescription is currently not available.\n\n\n * Cookie\nacq.sig\n * Duration\npast\n * Description\nDescription is currently not available.\n\n\n * Cookie\nptc\n * Duration\n2 years\n * Description\nNo description available.\n\n\nAdvertisement\nAdvertisement cookies are used to provide visitors with customised advertisements based on the pages you visited previously and to analyse the effectiveness of the ad campaigns.\n * Cookie\nmuc_ads\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to collect user behaviour and interaction data to optimize the website.\n\n\n * Cookie\nguest_id_marketing\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to identify and track the website visitor.\n\n\n * Cookie\nguest_id_ads\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to identify and track the website visitor.\n\n\n * Cookie\npersonalization_id\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to integrate and share features for social media and also store information about how the user uses the website, for tracking and targeting.\n\n\n * Cookie\nguest_id\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to identify and track the website visitor. It registers if a user is signed in to the Twitter platform and collects information about ad preferences.\n\n\n * Cookie\nbcookie\n * Duration\n1 year\n * Description\nLinkedIn sets this cookie from LinkedIn share buttons and ad tags to recognize browser IDs.\n\n\n * Cookie\n__Secure-ROLLOUT_TOKEN\n * Duration\n6 months\n * Description\nYouTube sets this cookie to manage feature rollout and experimentation. It helps Google control which new features or interface changes are shown to users as part of testing and staged rollouts, ensuring consistent experience for a given user during an experiment.\n\n\n * Cookie\nyt.innertube::nextId\n * Duration\nNever Expires\n * Description\nYouTube sets this cookie to register a unique ID to store data on what videos from YouTube the user has seen.\n\n\n * Cookie\nyt.innertube::requests\n * Duration\nNever Expires\n * Description\nYouTube sets this cookie to register a unique ID to store data on what videos from YouTube the user has seen.\n\n\n * Cookie\nsession_tracker\n * Duration\nsession\n * Description\nThis cookie is set by the Reddit. This cookie is used to identify trusted web traffic. It also helps in adverstising on the website.\n\n\n * Cookie\nedgebucket\n * Duration\nsession\n * Description\nReddit sets this cookie to save the information about a log-on Reddit user, for the purpose of advertisement recommendations and updating the content.\n\n\n * Cookie\ndid\n * Duration\n1 year\n * Description\nArbor sets this cookie to show targeted ads to site visitors.This cookie expires after 2 months or 1 year.\n\n\nUncategorised\nOther uncategorised cookies are those that are being analysed and have not been classified into a category as yet.\nNo cookies to display.\nReject All Save My Preferences Accept All\nToggle Search[](https://towardsdatascience.com/)Toggle Mobile Navigation\n[](https://towardsdatascience.com/)\nPublish AI, ML & data-science insights to a global community of data professionals.\nSign in[Submit an Article](https://write.insightmediagroup.io)\n[Latest](https://towardsdatascience.com/latest/)[Editor's Picks](https://towardsdatascience.com/tag/editors-pick/)[Newsletter](https://towardsdatascience.com/tag/the-variable/)[ShipAI](https://towardsdatascience.com/shipai/)\n* * *\n[Write For TDS](https://towardsdatascience.com/submissions/)\n[LinkedIn](https://www.linkedin.com/company/towards-data-science/)[X](https://x.com/TDataScience)Toggle Search\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\n# How LLMs Think\nResearch paper in pills: \"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"\n[Cristian Leo](https://towardsdatascience.com/author/cristianleo120/)\nJune 7, 202410 min read\n![Image Generated by DALL-E](../images/000.jpg)\nImage Generated by DALL-E\nHave you ever wondered how an AI model \"thinks\"? Imagine peering inside the mind of a machine and watching the gears turn. This is exactly what a groundbreaking paper from Anthropic explores. Titled \"[Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\", the research delves into understanding and interpreting the thought processes of AI.\nThe researchers managed to extract features from the Claude 3 Sonnet model that show what it was thinking about famous people, cities, and even security vulnerabilities in software. It's like getting a glimpse into the AI's mind, revealing the concepts it understands and uses to make decisions.\n### Research Paper Overview\nIn the paper, the Anthropic team, including Adly Templeton, Tom Conerly, Jonathan Marcus, and others, set out to make AI models more transparent. They focused on Claude 3 Sonnet, a medium-sized AI model, and aimed to scale monosemanticity - essentially making sure that each feature in the model has a clear, single meaning.\nBut why is scaling monosemanticity so important? And what exactly is monosemanticity? We'll dive into that soon.\nLearn this step by step with the interactive [AI Engineer](https://roadmap.sh/ai-engineer) and [Machine Learning](https://roadmap.sh/machine-learning) roadmaps.\n### Importance of the Study\nUnderstanding and interpreting features in AI models is crucial. It helps us see how these models make decisions, making them more reliable and easier to improve. When we can interpret these features, debugging, refining, and optimizing AI models becomes easier.\nThis research also has significant implications for AI safety. By identifying features linked to harmful behaviors, such as bias, deception, or dangerous content, we can develop ways to reduce these risks. This is especially important as AI systems become more integrated into everyday life, where ethical considerations and safety are essential.\nOne of the key contributions of this research is showing us how to understand what a large language model (LLM) is \"thinking.\" By extracting and interpreting features, we can get an insight into the internal workings of these complex models. This helps us see why they make certain decisions, providing a way to peek into their \"thought processes.\"\n### Background\nLet's review some of the odd terms mentioned earlier:\n**Monosemanticity** Monosemanticity is like having a single, specific key for each lock in a huge building. Imagine this building represents the AI model; each lock is a feature or concept the model understands. With monosemanticity, every key (feature) fits only one lock (concept) perfectly. This means whenever a particular key is used, it always opens the same lock. This consistency helps us understand exactly what the model is thinking about when it makes decisions because we know which key opened which lock.\n**Sparse Autoencoders** A sparse autoencoder is like a highly efficient detective. Imagine you have a big, cluttered room (the data) with many items scattered around. The detective's job is to find the few key items (important features) that tell the whole story of what happened in the room. The \"sparse\" part means this detective tries to solve the mystery using as few clues as possible, focusing only on the most essential pieces of evidence. In this research, sparse autoencoders act like this detective, helping to identify and extract clear, understandable features from the AI model, making it easier to see what's going on inside.\nHere are some useful [lecture notes by Andrew Ng on Autoencoders](http://stanford.edu/class/cs294a/sparseAutoencoder.pdf), to learn more about them.\n### Previous Work\nPrevious research laid the foundation by exploring how to extract interpretable features from smaller AI models using sparse autoencoders. These studies showed that sparse autoencoders could effectively identify meaningful features in simpler models. However, there were significant concerns about whether this method could scale up to larger, more complex models like Claude 3 Sonnet.\nThe earlier studies focused on proving that sparse autoencoders could identify and represent key features in smaller models. They succeeded in showing that the extracted features were both meaningful and interpretable. However, the main limitation was that these techniques had only been tested on simpler models. Scaling up was essential because larger models like Claude 3 Sonnet handle more complex data and tasks, making it harder to maintain the same level of clarity and usefulness in the extracted features.\nThis research builds on those foundations by aiming to scale these methods to more advanced AI systems. The researchers applied and adapted sparse autoencoders to handle the higher complexity and dimensionality of larger models. By addressing the challenges of scaling, this study seeks to ensure that even in more complex models, the extracted features remain clear and useful, thus advancing our understanding and interpretation of AI decision-making processes.\n### Scaling Sparse Autoencoders\nScaling sparse autoencoders to work with a larger model like Claude 3 Sonnet is like upgrading from a small, local library to managing a vast national archive. The techniques that worked well for the smaller collection need to be adjusted to handle the sheer size and complexity of the bigger dataset.\nSparse autoencoders are designed to identify and represent key features in data while keeping the number of active features low, much like a librarian who knows exactly which few books out of thousands will answer your question.\n![Image generated by DALL-E](../images/001.jpg)\nImage generated by DALL-E\nTwo key hypotheses guide this scaling:\n**Linear Representation Hypothesis** Imagine a giant map of the night sky, where each star represents a concept the AI understands. This hypothesis suggests that each concept (or star) aligns in a specific direction in the model's activation space. Essentially, it's like saying that if you draw a line through space pointing directly to a specific star, you can identify that star uniquely by its direction.\n**Superposition Hypothesis** Building on the night sky analogy, this hypothesis is like saying the AI can use these directions to map more stars than there are directions by using almost perpendicular lines. This allows the AI to efficiently pack information by finding unique ways to combine these directions, much like fitting more stars into the sky by carefully mapping them in different layers.\nBy applying these hypotheses, researchers could effectively scale sparse autoencoders to work with larger models like Claude 3 Sonnet, enabling them to capture and represent both simple and complex features in the data.\n### Training the Model\nImagine trying to train a group of detectives to sift through a vast library to find key pieces of evidence. This is similar to what researchers did with sparse autoencoders (SAEs) in their work with Claude 3 Sonnet, a complex AI model. They had to adapt the training techniques for these detectives to handle the larger, more complex data set represented by the Claude 3 Sonnet model.\nThe researchers decided to apply the SAEs to the residual stream activations in the middle layer of the model. Think of the middle layer as a crucial checkpoint in a detective's investigation, where a lot of interesting, abstract clues are found. They chose this point because:\n * **Smaller Size** : The residual stream is smaller than other layers, making it cheaper in terms of computational resources.\n * **Mitigating Cross-Layer Superposition** : This refers to the problem of signals from different layers getting mixed up, like flavors blending together in a way that makes it hard to tell them apart.\n * **Rich in Abstract Features** : The middle layer is likely to contain intriguing, high-level concepts.\n\n\nThe team trained three versions of the SAEs, with different capacities to handle features: 1M features, 4M features, and 34M features. For each SAE, the goal was to keep the number of active features low while maintaining accuracy:\n * **Active Features** : On average, fewer than 300 features were active at any time, explaining at least 65% of the variance in the model's activations.\n * **Dead Features** : These are features that never get activated. They found roughly 2% dead features in the 1M SAE, 35% in the 4M SAE, and 65% in the 34M SAE. Future improvements aim to reduce these numbers.\n\n\n### Scaling Laws: Optimizing Training\nThe goal was to balance reconstruction accuracy with the number of active features, using a loss function that combined mean-squared error (MSE) and an L1 penalty.\nAlso, they applied scaling laws, which help determine how many training steps and features are optimal within a given compute budget. Essentially, scaling laws tell us that as we increase our computing resources, the number of features and training steps should increase according to a predictable pattern, often following a power law.\nAs they increased the compute budget, the optimal number of features and training steps scaled according to a power law.\n![Loss by features and training steps plot - Image extracted by "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet"](../images/002.png)\nLoss by features and training steps plot - Image extracted by [\"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\nThey found that the best learning rates also followed a power law trend, helping them choose appropriate rates for larger runs.\n### Mathematical Foundation\nThe core mathematical principles behind the sparse autoencoder model are essential for understanding how it decomposes activations into interpretable features.\n**Encoder** The encoder transforms the input activations into a higher-dimensional space using a learned linear transformation followed by a ReLU nonlinearity. This is represented as:\n![Encoder Function - Image by Author](../images/003.png)\nEncoder Function - Image by Author\nHere, _W^_enc and _b^_enc are the encoder weights and biases, and _fi_ ​(_x_) represents the activation of feature _i_.\n**Decoder** The decoder attempts to reconstruct the original activations from the features using another linear transformation:\n![Decoder Function - Image by Author](../images/004.png)\nDecoder Function - Image by Author\n_W^_dec and _b^_dec are the decoder weights and biases. The term _fi_ ​(_x_)_W^_dec​ represents the contribution of feature _i_ to the reconstruction.\n**Loss** The model is trained to minimize a combination of reconstruction error and sparsity penalty:\n![Loss Function - Image by Author](../images/005.png)\nLoss Function - Image by Author\nThis loss function ensures that the reconstruction is accurate (minimizing the L2 norm of the error) while keeping the number of active features low (enforced by the L1 regularization term with a coefficient _λ_).\n### Interpretable Features\nThe research revealed a wide variety of interpretable features within the Claude 3 Sonnet model, encompassing both abstract and concrete concepts. These features provide insights into the model's internal processes and decision-making patterns.\n**Abstract Features** : These include high-level concepts that the model understands and uses to process information. Examples are themes like emotions, intentions, and broader categories such as science or technology.\n**Concrete Features** : These are more specific and tangible, such as names of famous people, geographical locations, or particular objects. These features can be directly linked to identifiable real-world entities.\nFor instance, the model has features that activate in response to mentions of well-known individuals. There might be a feature specifically for \"Albert Einstein\" that activates whenever the text refers to him or his work in physics. This feature helps the model make connections and generate contextually relevant information about Einstein.\n![Albert Einstein feature - Image extracted by "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet"](../images/006.png)\nAlbert Einstein feature - Image extracted by [\"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\nSimilarly, there are features that respond to references to cities, countries, and other geographical entities. For example, a feature for \"Paris\" might activate when the text talks about the Eiffel Tower, French culture, or events happening in the city. This helps the model understand and contextualize discussions about these places.\nThe model can also identify and activate features related to security vulnerabilities in code or systems. For example, there might be a feature that recognizes mentions of \"buffer overflow\" or \"SQL injection,\" which are common security issues in software development. This capability is crucial for applications involving cybersecurity, as it allows the model to detect and highlight potential risks.\n![Security Measures - Image extracted by "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet"](../images/007.png)\nSecurity Measures - Image extracted by [\"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\nFeatures related to biases were also identified, including those that detect racial, gender, or other forms of prejudice. By understanding these features, developers can work to mitigate biased outputs, ensuring that the AI behaves more fairly and equitably.\n![Gender Bias - Image extracted by "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet"](../images/008.png)\nGender Bias - Image extracted by [\"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\nThese interpretable features demonstrate the model's ability to capture and utilize both specific and broad concepts. By understanding these features, researchers can better grasp how Claude 3 Sonnet processes information, making the model's actions more transparent and predictable. This understanding is vital for improving AI reliability, safety, and alignment with human values.\n### Conclusion\nThis research has made significant strides in understanding and interpreting the internal workings of the Claude 3 Sonnet model.\nThe study successfully extracted both abstract and concrete features from Claude 3 Sonnet, making the AI's decision-making process more transparent. Examples include features for famous people, cities, and security vulnerabilities.\nThe research identified features related to AI safety, such as detecting security vulnerabilities, biases, and deceptive behaviors. Understanding these features is crucial for developing safer and more reliable AI systems.\nThe importance of interpretable AI features cannot be overstated. They enhance our ability to debug, refine, and optimize AI models, leading to better performance and reliability. Moreover, they are essential for ensuring AI systems operate transparently and align with human values, particularly in areas of safety and ethics.\n## References\n 1. **Anthropic. Adly Templeton et al.** \"[Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html).\" _Anthropic Research_ , 2024.\n 2. **Ng, Andrew.** \"[Autoencoders: Overview and Applications](https://web.stanford.edu/class/cs294a/sparseAutoencoder_2011new.pdf).\" _Lecture Notes_ , Stanford University.\n 3. **Anthropic.** \"[Core Views on AI Safety](https://www.anthropic.com/news/core-views-on-ai-safety).\" _Anthropic Safety Guidelines_ , 2024.\n\n\n## Register Now to Continue Reading\n### Unlock Towards Data Science with a Free Account\nAll articles on Towards Data Science are available to registered users. Create a free account today to access this article and thousands more.\nRegister for freeSign in to your account\n* * *\n![](../images/body-8f73ac5a40c91a7b503f.jpeg)\nWritten By\n## [Cristian Leo](https://towardsdatascience.com/author/cristianleo120/)\n[See all from Cristian Leo](https://towardsdatascience.com/author/cristianleo120/)\n[Artificial Intelligence](https://towardsdatascience.com/tag/artificial-intelligence/)[Getting Started](https://towardsdatascience.com/tag/getting-started/)[Large Language Models](https://towardsdatascience.com/tag/large-language-models/)[Machine Learning](https://towardsdatascience.com/tag/machine-learning/)[Monosemanticity](https://towardsdatascience.com/tag/monosemanticity/)\nShare This Article\n * [](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F&title=How%20LLMs%20Think)\n * [](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F&title=How%20LLMs%20Think)\n * [](https://x.com/share?url=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F&text=How%20LLMs%20Think)\n\n\nTowards Data Science is a community publication. Submit your insights to reach our global audience and earn through the TDS Author Payment Program.\n[Write for TDS](https://towardsdatascience.com/submissions/)\n## Related Articles\n * [![](../images/009.jpg)](https://towardsdatascience.com/an-overview-of-forecasting-performance-metrics-ef548dad0134/)\n## [An Overview Of Forecasting Performance Metrics](https://towardsdatascience.com/an-overview-of-forecasting-performance-metrics-ef548dad0134/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nAn overview of some of the fundamental performance metrics for time series forecasting\n[Egor Howell](https://towardsdatascience.com/author/egorhowell/)\nNovember 16, 20225 min read\n * [![](../images/010.jpg)](https://towardsdatascience.com/the-intuition-behind-transformers-attention-is-all-you-need-393b5cfb4ada/)\n## [The Intuition Behind Transformers - Attention is All You Need](https://towardsdatascience.com/the-intuition-behind-transformers-attention-is-all-you-need-393b5cfb4ada/)\n[Machine Learning](https://towardsdatascience.com/category/artificial-intelligence/machine-learning/)\nTraditionally recurrent neural networks and their variants have been used extensively for Natural Language Processing problems. In recent...\n[Sam Palani](https://towardsdatascience.com/author/samx18/)\nNovember 1, 20207 min read\n * [![](../images/011.jpeg)](https://towardsdatascience.com/an-utterly-simple-guide-on-installing-tensorflow-gpu-2-0-on-windows-10-198368dc07a1/)\n## [An Utterly Simple Guide on Installing Tensorflow-GPU 2.0 on Windows 10](https://towardsdatascience.com/an-utterly-simple-guide-on-installing-tensorflow-gpu-2-0-on-windows-10-198368dc07a1/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nShouldn't take you more than 15 minutes - depending on your Internet speed.\n[Dario Radečić](https://towardsdatascience.com/author/radecicdario/)\nFebruary 29, 20204 min read\n * [![](../images/012.jpg)](https://towardsdatascience.com/lasso-regression-why-the-solution-lives-on-a-diamond/)\n## [Lasso Regression: Why the Solution Lives on a Diamond](https://towardsdatascience.com/lasso-regression-why-the-solution-lives-on-a-diamond/)\n[Machine Learning](https://towardsdatascience.com/category/artificial-intelligence/machine-learning/)\nIt’s simpler than you think.\n[Nikhil Dasari](https://towardsdatascience.com/author/nikhil.dasari/)\nApril 23, 202620 min read\n * [![](../images/013.jpeg)](https://towardsdatascience.com/my-best-method-to-learn-new-tools-in-data-science-31941c2fc810/)\n## [My Best Method to Learn New Tools in Data Science](https://towardsdatascience.com/my-best-method-to-learn-new-tools-in-data-science-31941c2fc810/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nAnd how I practice it.\n[Soner Yıldırım](https://towardsdatascience.com/author/sonery/)\nFebruary 10, 20215 min read\n * [![](../images/014.jpg)](https://towardsdatascience.com/making-your-first-kaggle-submission-36fa07739272/)\n## [Making Your First Kaggle Submission](https://towardsdatascience.com/making-your-first-kaggle-submission-36fa07739272/)\n[Deep Learning](https://towardsdatascience.com/category/artificial-intelligence/deep-learning/)\nAn easy-to-understand guide to getting started with competitions and successfully modelling and making your first submission.\n[Yash Prakash](https://towardsdatascience.com/author/ipom/)\nMarch 21, 20217 min read\n * [![](../images/015.jpg)](https://towardsdatascience.com/debunking-data-science-myths-83d693fb5ba4/)\n## [Debunking Data Science Myths](https://towardsdatascience.com/debunking-data-science-myths-83d693fb5ba4/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nThe Truth They Don't Tell You About Data Science\n[Kurtis Pykes](https://towardsdatascience.com/author/kurtispykes/)\nDecember 8, 20204 min read\n * [![](../images/016.jpg)](https://towardsdatascience.com/python-programming-concepts-that-made-my-code-efficient-68f92f8a39d0/)\n## [Python programming concepts that made my code efficient!](https://towardsdatascience.com/python-programming-concepts-that-made-my-code-efficient-68f92f8a39d0/)\n[Machine Learning](https://towardsdatascience.com/category/artificial-intelligence/machine-learning/)\nUse these magic commands to empower your Python class!\n[Pratik Gandhi](https://towardsdatascience.com/author/pratikkgandhi/)\nAugust 31, 20205 min read\n * [![](../images/017.png)](https://towardsdatascience.com/review-of-reinforcement-learning-papers-14-70b6772a8100-2/)\n## [Review of Reinforcement Learning Papers #14](https://towardsdatascience.com/review-of-reinforcement-learning-papers-14-70b6772a8100-2/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nI present 4 publications from my research area: reinforcement learning. Let's discuss it!\n[Quentin Gallouédec](https://towardsdatascience.com/author/qgallouedec/)\nDecember 19, 20220 min read\n\n\n * [YouTube](https://www.youtube.com/c/TowardsDataScience)\n * [X](https://x.com/TDataScience)\n * [LinkedIn](https://www.linkedin.com/company/towards-data-science/)\n * [Threads](https://www.threads.net/@towardsdatascience)\n * [Bluesky](https://bsky.app/profile/towardsdatascience.com)\n\n\n[](https://towardsdatascience.com/)\nYour home for data science and AI. The world's leading publication for data science, data analytics, data engineering, machine learning, and artificial intelligence professionals.\nSubscribe to Our Newsletter[Write For TDS](https://towardsdatascience.com/submissions/)•[About](https://towardsdatascience.com/about-towards-data-science-d691af11cc2f/)•[Advertise](https://contact.towardsdatascience.com/advertise-with-towards-data-science)•[Privacy Policy](https://towardsdatascience.com/privacy-policy/)•[Terms of Use](https://towardsdatascience.com/website-terms-of-use/)\n© Insight Media Group, LLC 2026•[The New Stack](https://thenewstack.io)•[roadmap.sh](https://roadmap.sh)•[Creator Network](https://insightmediagroup.io/creator-network)\n#### Recommended Articles \nClose\n * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607686/image?etag=1782161464448&width=500&height=500) How Relevance Models Foreshadowed Transformers for NLP ](https://towardsdatascience.com/how-relevance-models-foreshadowed-transformers-for-nlp/ \"How Relevance Models Foreshadowed Transformers for NLP\")\n * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607639/image?etag=1781250125848&width=500&height=500) Javascript Fatigue: HTMX Is All You Need to Build ChatGPT — Part 1 ](https://towardsdatascience.com/javascript-fatigue-you-dont-need-js-to-build-chatgpt/ \"Javascript Fatigue: HTMX Is All You Need to Build ChatGPT — Part 1\")\n * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607930/image?etag=1782124566054&width=500&height=500) The Subset Sum Problem Solved in Linear Time for Dense Enough Inputs ](https://towardsdatascience.com/subset-sum-problem-solved-in-linear-time-for-dense-enough-inputs/ \"The Subset Sum Problem Solved in Linear Time for Dense Enough Inputs\")\n * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/3660/image?etag=1780176246070&width=500&height=500) Does Your Company Have a Data Strategy? ](https://towardsdatascience.com/does-your-company-have-a-data-strategy-e8944dbec4ae/ \"Does Your Company Have a Data Strategy?\")\n\n\n", "metadata": { "title": "How LLMs Think | Towards Data Science", "description": "Research paper in pills: \"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"", "keywords": null, "author": "Cristian Leo", "og:locale": "en_US", "og:type": "article", "og:title": "How LLMs Think", "og:description": "Research paper in pills: \"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"", "og:url": "https://towardsdatascience.com/how-llms-think-d8754a79017d/", "og:site_name": "Towards Data Science", "og:image": "https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/08C81qDB315xvb8d6.jpg", "og:image:width": "1792", "og:image:height": "1024", "og:image:type": "image/jpeg", "twitter:card": "summary_large_image", "twitter:site": "@TDataScience", "twitter:creator": "@TDataScience", "twitter:label1": "Written by", "twitter:data1": "Cristian Leo", "twitter:label2": "Est. reading time", "twitter:data2": "10 min read", "article:published_time": "2024-06-07T06:27:18.000Z", "article:modified_time": "2024-06-07T06:27:18.000Z", "article:author": "https://towardsdatascience.com/author/cristianleo120/" }, "paywall": { "listed_paid": false, "detected": false, "marker": "", "public_only": false, "policy": "付费/会员内容只保留页面公开可见部分，不绕过付费墙" }, "crawl": { "success": true, "status_code": 202, "error": "" }, "code_blocks": [], "images": [ { "status": "http_404", "url": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607686/image?etag=1782161464448&width=500&height=500", "src": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607686/image?etag=1782161464448&width=500&height=500", "alt": "", "downloaded": false, "discovery": "saved_markdown", "retried_at": "2026-10-20T18:06:17Z" }, { "status": "http_404", "url": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607639/image?etag=1781250125848&width=500&height=500", "src": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607639/image?etag=1781250125848&width=500&height=500", "alt": "", "downloaded": false, "discovery": "saved_markdown", "retried_at": "2026-10-20T18:06:17Z" }, { "status": "http_404", "url": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607930/image?etag=1782124566054&width=500&height=500", "src": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607930/image?etag=1782124566054&width=500&height=500", "alt": "", "downloaded": false, "discovery": "saved_markdown", "retried_at": "2026-10-20T18:06:17Z" }, { "status": "http_404", "url": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/3660/image?etag=1780176246070&width=500&height=500", "src": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/3660/image?etag=1780176246070&width=500&height=500", "alt": "", "downloaded": false, "discovery": "saved_markdown", "retried_at": "2026-10-20T18:06:17Z" } ], "audio": [], "videos": [ { "url": "https://www.youtube.com/c/TowardsDataScience", "kind": "channel_reference", "status": "reference_only" } ], "links": [ { "type": "internal", "href": "https://towardsdatascience.com/" }, { "type": "internal", "href": "https://towardsdatascience.com/latest/" }, { "type": "internal", "href": "https://towardsdatascience.com/tag/editors-pick/" }, { "type": "internal", "href": "https://towardsdatascience.com/tag/the-variable/" }, { "type": "internal", "href": "https://towardsdatascience.com/shipai/" }, { "type": "internal", "href": "https://towardsdatascience.com/submissions/" }, { "type": "internal", "href": "https://towardsdatascience.com/category/artificial-intelligence/" }, { "type": "internal", "href": "https://towardsdatascience.com/author/cristianleo120/" }, { "type": "internal", "href": "https://towardsdatascience.com/tag/artificial-intelligence/" }, { "type": "internal", "href": "https://towardsdatascience.com/tag/getting-started/" }, { "type": "internal", "href": "https://towardsdatascience.com/tag/large-language-models/" }, { "type": "internal", "href": "https://towardsdatascience.com/tag/machine-learning/" }, { "type": "internal", "href": "https://towardsdatascience.com/tag/monosemanticity/" }, { "type": "internal", "href": "https://towardsdatascience.com/an-overview-of-forecasting-performance-metrics-ef548dad0134/" }, { "type": "internal", "href": "https://towardsdatascience.com/author/egorhowell/" }, { "type": "internal", "href": "https://towardsdatascience.com/the-intuition-behind-transformers-attention-is-all-you-need-393b5cfb4ada/" }, { "type": "internal", "href": "https://towardsdatascience.com/category/artificial-intelligence/machine-learning/" }, { "type": "internal", "href": "https://towardsdatascience.com/author/samx18/" }, { "type": "internal", "href": "https://towardsdatascience.com/an-utterly-simple-guide-on-installing-tensorflow-gpu-2-0-on-windows-10-198368dc07a1/" }, { "type": "internal", "href": "https://towardsdatascience.com/author/radecicdario/" }, { "type": "external", "href": "https://write.insightmediagroup.io" }, { "type": "external", "href": "https://www.linkedin.com/company/towards-data-science/" }, { "type": "external", "href": "https://x.com/TDataScience" }, { "type": "external", "href": "https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html" }, { "type": "external", "href": "https://roadmap.sh/ai-engineer" }, { "type": "external", "href": "https://roadmap.sh/machine-learning" }, { "type": "external", "href": "http://stanford.edu/class/cs294a/sparseAutoencoder.pdf" }, { "type": "external", "href": "https://web.stanford.edu/class/cs294a/sparseAutoencoder_2011new.pdf" }, { "type": "external", "href": "https://www.anthropic.com/news/core-views-on-ai-safety" }, { "type": "external", "href": "https://www.facebook.com/sharer/sharer.php?title=How+LLMs+Think&u=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F" }, { "type": "external", "href": "https://www.linkedin.com/shareArticle?mini=true&title=How+LLMs+Think&url=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F" }, { "type": "external", "href": "https://x.com/share?text=How+LLMs+Think&url=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F" }, { "type": "external", "href": "https://www.youtube.com/c/TowardsDataScience" }, { "type": "external", "href": "https://www.threads.net/@towardsdatascience" }, { "type": "external", "href": "https://bsky.app/profile/towardsdatascience.com" }, { "type": "external", "href": "https://thenewstack.io" }, { "type": "external", "href": "https://roadmap.sh" }, { "type": "external", "href": "https://insightmediagroup.io/creator-network" } ], "repair_version": 2, "repaired_at": "2026-10-20T17:32:36Z", "source_url": "https://towardsdatascience.com/how-llms-think-d8754a79017d/", "delivery_status": "unverified", "discovery_status": "http_403", "missing_reason": "源页面未重新核验:http_403；已按保留正文补采图片", "repair_error": "", "actual_video_files": [], "expected_saved_body_images": 4, "image_discovery": "saved_markdown", "source_body_markdown": "![Revisit consent button](https://cdn-cookieyes.com/assets/images/revisit.svg)\nWe value your privacy\nWe use cookies to enhance your browsing experience, serve personalised ads or content, and analyse our traffic. By clicking \"Accept All\", you consent to our use of cookies.\nCustomise Reject All Accept All\nCustomise Consent Preferences ![](https://cdn-cookieyes.com/assets/images/close.svg)\nWe use cookies to help you navigate efficiently and perform certain functions. You will find detailed information about all cookies under each consent category below.\nThe cookies that are categorised as \"Necessary\" are stored on your browser as they are essential for enabling the basic functionalities of the site. ... Show more\nNecessaryAlways Active\nNecessary cookies are required to enable the basic features of this site, such as providing secure log-in or adjusting your consent preferences. These cookies do not store any personally identifiable data.\n * Cookie\nBCTempID\n * Duration\n10 minutes\n * Description\nNo description available.\n\n\n * Cookie\n__cf_bm\n * Duration\n1 hour\n * Description\nThis cookie, set by Cloudflare, is used to support Cloudflare Bot Management. \n\n\n * Cookie\nAWSALBCORS\n * Duration\n7 days\n * Description\nAmazon Web Services set this cookie for load balancing.\n\n\n * Cookie\n_cfuvid\n * Duration\nsession\n * Description\nCloudflare sets this cookie to track users across sessions to optimize user experience by maintaining session consistency and providing personalized services\n\n\n * Cookie\nli_gc\n * Duration\n6 months\n * Description\nLinkedin set this cookie for storing visitor's consent regarding using cookies for non-essential purposes.\n\n\n * Cookie\n__hssrc\n * Duration\nsession\n * Description\nThis cookie is set by Hubspot whenever it changes the session cookie. The __hssrc cookie set to 1 indicates that the user has restarted the browser, and if the cookie does not exist, it is assumed to be a new session.\n\n\n * Cookie\n__hssc\n * Duration\n1 hour\n * Description\nHubSpot sets this cookie to keep track of sessions and to determine if HubSpot should increment the session number and timestamps in the __hstc cookie.\n\n\n * Cookie\nwpEmojiSettingsSupports\n * Duration\nsession\n * Description\nWordPress sets this cookie when a user interacts with emojis on a WordPress site. It helps determine if the user's browser can display emojis properly.\n\n\n * Cookie\nBCSessionID\n * Duration\n1 year 1 month 4 days\n * Description\nBlueconic sets this cookie as a unique identifier for the BlueConic profile.\n\n\n * Cookie\n_octo\n * Duration\n1 year\n * Description\nNo description available.\n\n\n * Cookie\nlogged_in\n * Duration\n1 year\n * Description\nNo description available.\n\n\n * Cookie\n__Secure-YEC\n * Duration\npast\n * Description\nYouTube sets this cookie to stores the user's video player preferences using embedded YouTube video\n\n\n * Cookie\n__eoi\n * Duration\n6 months\n * Description\nDescription is currently not available.\n\n\n * Cookie\nAWSALBTGCORS\n * Duration\n7 days\n * Description\nNo description available.\n\n\n * Cookie\nlogin-status-p\n * Duration\npast\n * Description\nDescription is currently not available.\n\n\n * Cookie\nAWSALBTG\n * Duration\n7 days\n * Description\nNo description available.\n\n\n * Cookie\ncsrf_token\n * Duration\nsession\n * Description\nNo description available.\n\n\n * Cookie\ntoken_v2\n * Duration\n1 day\n * Description\nDescription is currently not available.\n\n\n * Cookie\nD\n * Duration\n1 year\n * Description\nDescription is currently not available.\n\n\n * Cookie\nPHPSESSID\n * Duration\nsession\n * Description\nThis cookie is native to PHP applications. The cookie stores and identifies a user's unique session ID to manage user sessions on the website. The cookie is a session cookie and will be deleted when all the browser windows are closed.\n\n\n * Cookie\nVISITOR_PRIVACY_METADATA\n * Duration\n6 months\n * Description\nYouTube sets this cookie to store the user's cookie consent state for the current domain. \n\n\n * Cookie\ncookietest\n * Duration\nsession\n * Description\nThe cookietest cookie is typically used to determine whether the user's browser accepts cookies, essential for website functionality and user experience.\n\n\n * Cookie\n__Host-airtable-session\n * Duration\n1 year\n * Description\nThis cookie is used to enable us to integrate the services of Airtable.\n\n\n * Cookie\n__Host-airtable-session.sig\n * Duration\n1 year\n * Description\nThis cookie is used to enable us to integrate the services of Airtable.\n\n\n * Cookie\nm\n * Duration\n1 year 1 month 4 days\n * Description\nStripe sets this cookie for fraud prevention purposes. It identifies the device used to access the website, allowing the website to be formatted accordingly.\n\n\n * Cookie\nBIGipServer*\n * Duration\nsession\n * Description\nMarketo sets this cookie to collect information about the user's online activity and build a profile about their interests to provide advertisements relevant to the user. \n\n\n * Cookie\n__cfruid\n * Duration\nsession\n * Description\nCloudflare sets this cookie to identify trusted web traffic.\n\n\n * Cookie\n_GRECAPTCHA\n * Duration\n6 months\n * Description\nGoogle Recaptcha service sets this cookie to identify bots to protect the website against malicious spam attacks.\n\n\n * Cookie\n__Secure-YNID\n * Duration\n6 months\n * Description\nGoogle cookie used to protect user security and prevent fraud, especially during the login process.\n\n\n * Cookie\ncookieyes-consent\n * Duration\n1 year\n * Description\nCookieYes sets this cookie to remember users' consent preferences so that their preferences are respected on subsequent visits to this site. It does not collect or store any personal information about the site visitors.\n\n\nFunctional\nFunctional cookies help perform certain functionalities like sharing the content of the website on social media platforms, collecting feedback, and other third-party features.\n * Cookie\nlidc\n * Duration\n1 day\n * Description\nLinkedIn sets the lidc cookie to facilitate data center selection.\n\n\n * Cookie\nbrw\n * Duration\n1 year\n * Description\nNo description available.\n\n\n * Cookie\nbrwConsent\n * Duration\n5 minutes\n * Description\nDescription is currently not available.\n\n\n * Cookie\nWMF-Uniq\n * Duration\n1 year\n * Description\nDescription is currently not available.\n\n\n * Cookie\nloom_anon_comment\n * Duration\n1 year\n * Description\nNo description available.\n\n\n * Cookie\nloom_referral_video\n * Duration\nsession\n * Description\nDescription is currently not available.\n\n\n * Cookie\nVISITOR_INFO1_LIVE\n * Duration\n6 months\n * Description\nA cookie set by YouTube to measure bandwidth that determines whether the user gets the new or old player interface.\n\n\n * Cookie\nyt-remote-connected-devices\n * Duration\nNever Expires\n * Description\nYouTube sets this cookie to store the user's video preferences using embedded YouTube videos.\n\n\n * Cookie\nytidb::LAST_RESULT_ENTRY_KEY\n * Duration\nNever Expires\n * Description\nThe cookie ytidb::LAST_RESULT_ENTRY_KEY is used by YouTube to store the last search result entry that was clicked by the user. This information is used to improve the user experience by providing more relevant search results in the future.\n\n\n * Cookie\nyt-remote-device-id\n * Duration\nNever Expires\n * Description\nYouTube sets this cookie to store the user's video preferences using embedded YouTube videos.\n\n\n * Cookie\nyt-remote-session-name\n * Duration\nsession\n * Description\nThe yt-remote-session-name cookie is used by YouTube to store the user's video player preferences using embedded YouTube video.\n\n\n * Cookie\nyt-remote-fast-check-period\n * Duration\nsession\n * Description\nThe yt-remote-fast-check-period cookie is used by YouTube to store the user's video player preferences for embedded YouTube videos.\n\n\n * Cookie\nyt-remote-session-app\n * Duration\nsession\n * Description\nThe yt-remote-session-app cookie is used by YouTube to store user preferences and information about the interface of the embedded YouTube video player.\n\n\n * Cookie\nyt-remote-cast-available\n * Duration\nsession\n * Description\nThe yt-remote-cast-available cookie is used to store the user's preferences regarding whether casting is available on their YouTube video player.\n\n\n * Cookie\nyt-remote-cast-installed\n * Duration\nsession\n * Description\nThe yt-remote-cast-installed cookie is used to store the user's video player preferences using embedded YouTube video.\n\n\n * Cookie\ncp_session\n * Duration\n3 months\n * Description\nCodepen sets this cookie for Help systems found in the website.\n\n\n * Cookie\nloid\n * Duration\n1 year 1 month 4 days\n * Description\nThis cookie is set by the Reddit. The cookie enables the sharing of content from the website onto the social media platform.\n\n\nAnalytics\nAnalytical cookies are used to understand how visitors interact with the website. These cookies help provide information on metrics such as the number of visitors, bounce rate, traffic source, etc.\n * Cookie\n__hstc\n * Duration\n6 months\n * Description\nHubspot set this main cookie for tracking visitors. It contains the domain, initial timestamp (first visit), last timestamp (last visit), current timestamp (this visit), and session number (increments for each subsequent session).\n\n\n * Cookie\nhubspotutk\n * Duration\n6 months\n * Description\nHubSpot sets this cookie to keep track of the visitors to the website. This cookie is passed to HubSpot on form submission and used when deduplicating contacts.\n\n\n * Cookie\n_ga\n * Duration\n1 year 1 month 4 days\n * Description\nGoogle Analytics sets this cookie to calculate visitor, session and campaign data and track site usage for the site's analytics report. The cookie stores information anonymously and assigns a randomly generated number to recognise unique visitors.\n\n\n * Cookie\n_ga_*\n * Duration\n1 year 1 month 4 days\n * Description\nGoogle Analytics sets this cookie to store and count page views.\n\n\n * Cookie\n__Host-psifi.analyticsTrace\n * Duration\n6 hours\n * Description\nDescription is currently not available.\n\n\n * Cookie\n__Host-psifi.analyticsTraceV2\n * Duration\n6 hours\n * Description\nDescription is currently not available.\n\n\n * Cookie\n_gh_sess\n * Duration\nsession\n * Description\nGitHub sets this cookie for temporary application and framework state between pages like what step the user is on in a multiple step form.\n\n\n * Cookie\nYSC\n * Duration\nsession\n * Description\nYSC cookie is set by Youtube and is used to track the views of embedded videos on Youtube pages.\n\n\n * Cookie\najs_anonymous_id\n * Duration\n1 year\n * Description\nThis cookie is set by Segment to count the number of people who visit a certain site by tracking if they have visited before.\n\n\n * Cookie\nvuid\n * Duration\n1 year 1 month 4 days\n * Description\nVimeo installs this cookie to collect tracking information by setting a unique ID to embed videos on the website. \n\n\nPerformance\nPerformance cookies are used to understand and analyse the key performance indexes of the website which helps in delivering a better user experience for the visitors.\n * Cookie\nAWSALB\n * Duration\n7 days\n * Description\nAWSALB is an application load balancer cookie set by Amazon Web Services to map the session to the target.\n\n\n * Cookie\nacq\n * Duration\npast\n * Description\nDescription is currently not available.\n\n\n * Cookie\nacq.sig\n * Duration\npast\n * Description\nDescription is currently not available.\n\n\n * Cookie\nptc\n * Duration\n2 years\n * Description\nNo description available.\n\n\nAdvertisement\nAdvertisement cookies are used to provide visitors with customised advertisements based on the pages you visited previously and to analyse the effectiveness of the ad campaigns.\n * Cookie\nmuc_ads\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to collect user behaviour and interaction data to optimize the website.\n\n\n * Cookie\nguest_id_marketing\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to identify and track the website visitor.\n\n\n * Cookie\nguest_id_ads\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to identify and track the website visitor.\n\n\n * Cookie\npersonalization_id\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to integrate and share features for social media and also store information about how the user uses the website, for tracking and targeting.\n\n\n * Cookie\nguest_id\n * Duration\n1 year 1 month 4 days\n * Description\nTwitter sets this cookie to identify and track the website visitor. It registers if a user is signed in to the Twitter platform and collects information about ad preferences.\n\n\n * Cookie\nbcookie\n * Duration\n1 year\n * Description\nLinkedIn sets this cookie from LinkedIn share buttons and ad tags to recognize browser IDs.\n\n\n * Cookie\n__Secure-ROLLOUT_TOKEN\n * Duration\n6 months\n * Description\nYouTube sets this cookie to manage feature rollout and experimentation. It helps Google control which new features or interface changes are shown to users as part of testing and staged rollouts, ensuring consistent experience for a given user during an experiment.\n\n\n * Cookie\nyt.innertube::nextId\n * Duration\nNever Expires\n * Description\nYouTube sets this cookie to register a unique ID to store data on what videos from YouTube the user has seen.\n\n\n * Cookie\nyt.innertube::requests\n * Duration\nNever Expires\n * Description\nYouTube sets this cookie to register a unique ID to store data on what videos from YouTube the user has seen.\n\n\n * Cookie\nsession_tracker\n * Duration\nsession\n * Description\nThis cookie is set by the Reddit. This cookie is used to identify trusted web traffic. It also helps in adverstising on the website.\n\n\n * Cookie\nedgebucket\n * Duration\nsession\n * Description\nReddit sets this cookie to save the information about a log-on Reddit user, for the purpose of advertisement recommendations and updating the content.\n\n\n * Cookie\ndid\n * Duration\n1 year\n * Description\nArbor sets this cookie to show targeted ads to site visitors.This cookie expires after 2 months or 1 year.\n\n\nUncategorised\nOther uncategorised cookies are those that are being analysed and have not been classified into a category as yet.\nNo cookies to display.\nReject All Save My Preferences Accept All\nToggle Search[](https://towardsdatascience.com/)Toggle Mobile Navigation\n[](https://towardsdatascience.com/)\nPublish AI, ML & data-science insights to a global community of data professionals.\nSign in[Submit an Article](https://write.insightmediagroup.io)\n[Latest](https://towardsdatascience.com/latest/)[Editor's Picks](https://towardsdatascience.com/tag/editors-pick/)[Newsletter](https://towardsdatascience.com/tag/the-variable/)[ShipAI](https://towardsdatascience.com/shipai/)\n* * *\n[Write For TDS](https://towardsdatascience.com/submissions/)\n[LinkedIn](https://www.linkedin.com/company/towards-data-science/)[X](https://x.com/TDataScience)Toggle Search\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\n# How LLMs Think\nResearch paper in pills: \"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"\n[Cristian Leo](https://towardsdatascience.com/author/cristianleo120/)\nJune 7, 202410 min read\n![Image Generated by DALL-E](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/08C81qDB315xvb8d6.jpg)\nImage Generated by DALL-E\nHave you ever wondered how an AI model \"thinks\"? Imagine peering inside the mind of a machine and watching the gears turn. This is exactly what a groundbreaking paper from Anthropic explores. Titled \"[Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\", the research delves into understanding and interpreting the thought processes of AI.\nThe researchers managed to extract features from the Claude 3 Sonnet model that show what it was thinking about famous people, cities, and even security vulnerabilities in software. It's like getting a glimpse into the AI's mind, revealing the concepts it understands and uses to make decisions.\n### Research Paper Overview\nIn the paper, the Anthropic team, including Adly Templeton, Tom Conerly, Jonathan Marcus, and others, set out to make AI models more transparent. They focused on Claude 3 Sonnet, a medium-sized AI model, and aimed to scale monosemanticity - essentially making sure that each feature in the model has a clear, single meaning.\nBut why is scaling monosemanticity so important? And what exactly is monosemanticity? We'll dive into that soon.\nLearn this step by step with the interactive [AI Engineer](https://roadmap.sh/ai-engineer) and [Machine Learning](https://roadmap.sh/machine-learning) roadmaps.\n### Importance of the Study\nUnderstanding and interpreting features in AI models is crucial. It helps us see how these models make decisions, making them more reliable and easier to improve. When we can interpret these features, debugging, refining, and optimizing AI models becomes easier.\nThis research also has significant implications for AI safety. By identifying features linked to harmful behaviors, such as bias, deception, or dangerous content, we can develop ways to reduce these risks. This is especially important as AI systems become more integrated into everyday life, where ethical considerations and safety are essential.\nOne of the key contributions of this research is showing us how to understand what a large language model (LLM) is \"thinking.\" By extracting and interpreting features, we can get an insight into the internal workings of these complex models. This helps us see why they make certain decisions, providing a way to peek into their \"thought processes.\"\n### Background\nLet's review some of the odd terms mentioned earlier:\n**Monosemanticity** Monosemanticity is like having a single, specific key for each lock in a huge building. Imagine this building represents the AI model; each lock is a feature or concept the model understands. With monosemanticity, every key (feature) fits only one lock (concept) perfectly. This means whenever a particular key is used, it always opens the same lock. This consistency helps us understand exactly what the model is thinking about when it makes decisions because we know which key opened which lock.\n**Sparse Autoencoders** A sparse autoencoder is like a highly efficient detective. Imagine you have a big, cluttered room (the data) with many items scattered around. The detective's job is to find the few key items (important features) that tell the whole story of what happened in the room. The \"sparse\" part means this detective tries to solve the mystery using as few clues as possible, focusing only on the most essential pieces of evidence. In this research, sparse autoencoders act like this detective, helping to identify and extract clear, understandable features from the AI model, making it easier to see what's going on inside.\nHere are some useful [lecture notes by Andrew Ng on Autoencoders](http://stanford.edu/class/cs294a/sparseAutoencoder.pdf), to learn more about them.\n### Previous Work\nPrevious research laid the foundation by exploring how to extract interpretable features from smaller AI models using sparse autoencoders. These studies showed that sparse autoencoders could effectively identify meaningful features in simpler models. However, there were significant concerns about whether this method could scale up to larger, more complex models like Claude 3 Sonnet.\nThe earlier studies focused on proving that sparse autoencoders could identify and represent key features in smaller models. They succeeded in showing that the extracted features were both meaningful and interpretable. However, the main limitation was that these techniques had only been tested on simpler models. Scaling up was essential because larger models like Claude 3 Sonnet handle more complex data and tasks, making it harder to maintain the same level of clarity and usefulness in the extracted features.\nThis research builds on those foundations by aiming to scale these methods to more advanced AI systems. The researchers applied and adapted sparse autoencoders to handle the higher complexity and dimensionality of larger models. By addressing the challenges of scaling, this study seeks to ensure that even in more complex models, the extracted features remain clear and useful, thus advancing our understanding and interpretation of AI decision-making processes.\n### Scaling Sparse Autoencoders\nScaling sparse autoencoders to work with a larger model like Claude 3 Sonnet is like upgrading from a small, local library to managing a vast national archive. The techniques that worked well for the smaller collection need to be adjusted to handle the sheer size and complexity of the bigger dataset.\nSparse autoencoders are designed to identify and represent key features in data while keeping the number of active features low, much like a librarian who knows exactly which few books out of thousands will answer your question.\n![Image generated by DALL-E](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/0Mg2AnbBsuZVmbDzF-1.jpg)\nImage generated by DALL-E\nTwo key hypotheses guide this scaling:\n**Linear Representation Hypothesis** Imagine a giant map of the night sky, where each star represents a concept the AI understands. This hypothesis suggests that each concept (or star) aligns in a specific direction in the model's activation space. Essentially, it's like saying that if you draw a line through space pointing directly to a specific star, you can identify that star uniquely by its direction.\n**Superposition Hypothesis** Building on the night sky analogy, this hypothesis is like saying the AI can use these directions to map more stars than there are directions by using almost perpendicular lines. This allows the AI to efficiently pack information by finding unique ways to combine these directions, much like fitting more stars into the sky by carefully mapping them in different layers.\nBy applying these hypotheses, researchers could effectively scale sparse autoencoders to work with larger models like Claude 3 Sonnet, enabling them to capture and represent both simple and complex features in the data.\n### Training the Model\nImagine trying to train a group of detectives to sift through a vast library to find key pieces of evidence. This is similar to what researchers did with sparse autoencoders (SAEs) in their work with Claude 3 Sonnet, a complex AI model. They had to adapt the training techniques for these detectives to handle the larger, more complex data set represented by the Claude 3 Sonnet model.\nThe researchers decided to apply the SAEs to the residual stream activations in the middle layer of the model. Think of the middle layer as a crucial checkpoint in a detective's investigation, where a lot of interesting, abstract clues are found. They chose this point because:\n * **Smaller Size** : The residual stream is smaller than other layers, making it cheaper in terms of computational resources.\n * **Mitigating Cross-Layer Superposition** : This refers to the problem of signals from different layers getting mixed up, like flavors blending together in a way that makes it hard to tell them apart.\n * **Rich in Abstract Features** : The middle layer is likely to contain intriguing, high-level concepts.\n\n\nThe team trained three versions of the SAEs, with different capacities to handle features: 1M features, 4M features, and 34M features. For each SAE, the goal was to keep the number of active features low while maintaining accuracy:\n * **Active Features** : On average, fewer than 300 features were active at any time, explaining at least 65% of the variance in the model's activations.\n * **Dead Features** : These are features that never get activated. They found roughly 2% dead features in the 1M SAE, 35% in the 4M SAE, and 65% in the 34M SAE. Future improvements aim to reduce these numbers.\n\n\n### Scaling Laws: Optimizing Training\nThe goal was to balance reconstruction accuracy with the number of active features, using a loss function that combined mean-squared error (MSE) and an L1 penalty.\nAlso, they applied scaling laws, which help determine how many training steps and features are optimal within a given compute budget. Essentially, scaling laws tell us that as we increase our computing resources, the number of features and training steps should increase according to a predictable pattern, often following a power law.\nAs they increased the compute budget, the optimal number of features and training steps scaled according to a power law.\n![Loss by features and training steps plot - Image extracted by "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet"](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/1q4qiBG0EKr63zUD2kxzy5w-1.png)\nLoss by features and training steps plot - Image extracted by [\"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\nThey found that the best learning rates also followed a power law trend, helping them choose appropriate rates for larger runs.\n### Mathematical Foundation\nThe core mathematical principles behind the sparse autoencoder model are essential for understanding how it decomposes activations into interpretable features.\n**Encoder** The encoder transforms the input activations into a higher-dimensional space using a learned linear transformation followed by a ReLU nonlinearity. This is represented as:\n![Encoder Function - Image by Author](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/1osv9TS4_-JAWp8lu3Eg_jA-1.png)\nEncoder Function - Image by Author\nHere, _W^_enc and _b^_enc are the encoder weights and biases, and _fi_ ​(_x_) represents the activation of feature _i_.\n**Decoder** The decoder attempts to reconstruct the original activations from the features using another linear transformation:\n![Decoder Function - Image by Author](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/1zPjyGhse7RIthEyzXNr5bA-1.png)\nDecoder Function - Image by Author\n_W^_dec and _b^_dec are the decoder weights and biases. The term _fi_ ​(_x_)_W^_dec​ represents the contribution of feature _i_ to the reconstruction.\n**Loss** The model is trained to minimize a combination of reconstruction error and sparsity penalty:\n![Loss Function - Image by Author](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/1GGXWlOgicyhAQLzwRzEAcQ-1.png)\nLoss Function - Image by Author\nThis loss function ensures that the reconstruction is accurate (minimizing the L2 norm of the error) while keeping the number of active features low (enforced by the L1 regularization term with a coefficient _λ_).\n### Interpretable Features\nThe research revealed a wide variety of interpretable features within the Claude 3 Sonnet model, encompassing both abstract and concrete concepts. These features provide insights into the model's internal processes and decision-making patterns.\n**Abstract Features** : These include high-level concepts that the model understands and uses to process information. Examples are themes like emotions, intentions, and broader categories such as science or technology.\n**Concrete Features** : These are more specific and tangible, such as names of famous people, geographical locations, or particular objects. These features can be directly linked to identifiable real-world entities.\nFor instance, the model has features that activate in response to mentions of well-known individuals. There might be a feature specifically for \"Albert Einstein\" that activates whenever the text refers to him or his work in physics. This feature helps the model make connections and generate contextually relevant information about Einstein.\n![Albert Einstein feature - Image extracted by "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet"](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/142cB7SrzgIeKc9UVVx1QYg-1.png)\nAlbert Einstein feature - Image extracted by [\"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\nSimilarly, there are features that respond to references to cities, countries, and other geographical entities. For example, a feature for \"Paris\" might activate when the text talks about the Eiffel Tower, French culture, or events happening in the city. This helps the model understand and contextualize discussions about these places.\nThe model can also identify and activate features related to security vulnerabilities in code or systems. For example, there might be a feature that recognizes mentions of \"buffer overflow\" or \"SQL injection,\" which are common security issues in software development. This capability is crucial for applications involving cybersecurity, as it allows the model to detect and highlight potential risks.\n![Security Measures - Image extracted by "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet"](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/1JiAhy-SQ4EdiKP7T8yPpCQ-1.png)\nSecurity Measures - Image extracted by [\"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\nFeatures related to biases were also identified, including those that detect racial, gender, or other forms of prejudice. By understanding these features, developers can work to mitigate biased outputs, ensuring that the AI behaves more fairly and equitably.\n![Gender Bias - Image extracted by "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet"](https://assets.insightmediagroup.io/media/wp-content/uploads/2024/06/1HTdn9OubqiE5dvGiLl9HZA-1.png)\nGender Bias - Image extracted by [\"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet\"](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)\nThese interpretable features demonstrate the model's ability to capture and utilize both specific and broad concepts. By understanding these features, researchers can better grasp how Claude 3 Sonnet processes information, making the model's actions more transparent and predictable. This understanding is vital for improving AI reliability, safety, and alignment with human values.\n### Conclusion\nThis research has made significant strides in understanding and interpreting the internal workings of the Claude 3 Sonnet model.\nThe study successfully extracted both abstract and concrete features from Claude 3 Sonnet, making the AI's decision-making process more transparent. Examples include features for famous people, cities, and security vulnerabilities.\nThe research identified features related to AI safety, such as detecting security vulnerabilities, biases, and deceptive behaviors. Understanding these features is crucial for developing safer and more reliable AI systems.\nThe importance of interpretable AI features cannot be overstated. They enhance our ability to debug, refine, and optimize AI models, leading to better performance and reliability. Moreover, they are essential for ensuring AI systems operate transparently and align with human values, particularly in areas of safety and ethics.\n## References\n 1. **Anthropic. Adly Templeton et al.** \"[Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html).\" _Anthropic Research_ , 2024.\n 2. **Ng, Andrew.** \"[Autoencoders: Overview and Applications](https://web.stanford.edu/class/cs294a/sparseAutoencoder_2011new.pdf).\" _Lecture Notes_ , Stanford University.\n 3. **Anthropic.** \"[Core Views on AI Safety](https://www.anthropic.com/news/core-views-on-ai-safety).\" _Anthropic Safety Guidelines_ , 2024.\n\n\n## Register Now to Continue Reading\n### Unlock Towards Data Science with a Free Account\nAll articles on Towards Data Science are available to registered users. Create a free account today to access this article and thousands more.\nRegister for freeSign in to your account\n* * *\n![](https://assets.insightmediagroup.io/media/1787657505387_wlyc0o.jpeg)\nWritten By\n## [Cristian Leo](https://towardsdatascience.com/author/cristianleo120/)\n[See all from Cristian Leo](https://towardsdatascience.com/author/cristianleo120/)\n[Artificial Intelligence](https://towardsdatascience.com/tag/artificial-intelligence/)[Getting Started](https://towardsdatascience.com/tag/getting-started/)[Large Language Models](https://towardsdatascience.com/tag/large-language-models/)[Machine Learning](https://towardsdatascience.com/tag/machine-learning/)[Monosemanticity](https://towardsdatascience.com/tag/monosemanticity/)\nShare This Article\n * [](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F&title=How%20LLMs%20Think)\n * [](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F&title=How%20LLMs%20Think)\n * [](https://x.com/share?url=https%3A%2F%2Ftowardsdatascience.com%2Fhow-llms-think-d8754a79017d%2F&text=How%20LLMs%20Think)\n\n\nTowards Data Science is a community publication. Submit your insights to reach our global audience and earn through the TDS Author Payment Program.\n[Write for TDS](https://towardsdatascience.com/submissions/)\n## Related Articles\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2022/11/0ramo1Mj_UNrRQFdL-scaled.jpg)](https://towardsdatascience.com/an-overview-of-forecasting-performance-metrics-ef548dad0134/)\n## [An Overview Of Forecasting Performance Metrics](https://towardsdatascience.com/an-overview-of-forecasting-performance-metrics-ef548dad0134/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nAn overview of some of the fundamental performance metrics for time series forecasting\n[Egor Howell](https://towardsdatascience.com/author/egorhowell/)\nNovember 16, 20225 min read\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2020/11/0M-eueiSQSSKJ178v-scaled.jpg)](https://towardsdatascience.com/the-intuition-behind-transformers-attention-is-all-you-need-393b5cfb4ada/)\n## [The Intuition Behind Transformers - Attention is All You Need](https://towardsdatascience.com/the-intuition-behind-transformers-attention-is-all-you-need-393b5cfb4ada/)\n[Machine Learning](https://towardsdatascience.com/category/artificial-intelligence/machine-learning/)\nTraditionally recurrent neural networks and their variants have been used extensively for Natural Language Processing problems. In recent...\n[Sam Palani](https://towardsdatascience.com/author/samx18/)\nNovember 1, 20207 min read\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2020/02/1E4AoT2c-LVx421XAx5XdrA.jpeg)](https://towardsdatascience.com/an-utterly-simple-guide-on-installing-tensorflow-gpu-2-0-on-windows-10-198368dc07a1/)\n## [An Utterly Simple Guide on Installing Tensorflow-GPU 2.0 on Windows 10](https://towardsdatascience.com/an-utterly-simple-guide-on-installing-tensorflow-gpu-2-0-on-windows-10-198368dc07a1/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nShouldn't take you more than 15 minutes - depending on your Internet speed.\n[Dario Radečić](https://towardsdatascience.com/author/radecicdario/)\nFebruary 29, 20204 min read\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2026/04/pexels-brett-sayles-4178723-scaled-1.jpg)](https://towardsdatascience.com/lasso-regression-why-the-solution-lives-on-a-diamond/)\n## [Lasso Regression: Why the Solution Lives on a Diamond](https://towardsdatascience.com/lasso-regression-why-the-solution-lives-on-a-diamond/)\n[Machine Learning](https://towardsdatascience.com/category/artificial-intelligence/machine-learning/)\nIt’s simpler than you think.\n[Nikhil Dasari](https://towardsdatascience.com/author/nikhil.dasari/)\nApril 23, 202620 min read\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2021/02/1eJn9eEX7Lx8UCfRf2tgW0g-scaled.jpeg)](https://towardsdatascience.com/my-best-method-to-learn-new-tools-in-data-science-31941c2fc810/)\n## [My Best Method to Learn New Tools in Data Science](https://towardsdatascience.com/my-best-method-to-learn-new-tools-in-data-science-31941c2fc810/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nAnd how I practice it.\n[Soner Yıldırım](https://towardsdatascience.com/author/sonery/)\nFebruary 10, 20215 min read\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2021/03/0zUFQGHSBdEakePcO-scaled.jpg)](https://towardsdatascience.com/making-your-first-kaggle-submission-36fa07739272/)\n## [Making Your First Kaggle Submission](https://towardsdatascience.com/making-your-first-kaggle-submission-36fa07739272/)\n[Deep Learning](https://towardsdatascience.com/category/artificial-intelligence/deep-learning/)\nAn easy-to-understand guide to getting started with competitions and successfully modelling and making your first submission.\n[Yash Prakash](https://towardsdatascience.com/author/ipom/)\nMarch 21, 20217 min read\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2020/12/0M4gj9drSWPum4oBz-scaled.jpg)](https://towardsdatascience.com/debunking-data-science-myths-83d693fb5ba4/)\n## [Debunking Data Science Myths](https://towardsdatascience.com/debunking-data-science-myths-83d693fb5ba4/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nThe Truth They Don't Tell You About Data Science\n[Kurtis Pykes](https://towardsdatascience.com/author/kurtispykes/)\nDecember 8, 20204 min read\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2020/08/0yRhQ28Nd53cTSCR6-scaled.jpg)](https://towardsdatascience.com/python-programming-concepts-that-made-my-code-efficient-68f92f8a39d0/)\n## [Python programming concepts that made my code efficient!](https://towardsdatascience.com/python-programming-concepts-that-made-my-code-efficient-68f92f8a39d0/)\n[Machine Learning](https://towardsdatascience.com/category/artificial-intelligence/machine-learning/)\nUse these magic commands to empower your Python class!\n[Pratik Gandhi](https://towardsdatascience.com/author/pratikkgandhi/)\nAugust 31, 20205 min read\n * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2022/12/13UX5NuNmQ30wu_uBV5gWXw.png)](https://towardsdatascience.com/review-of-reinforcement-learning-papers-14-70b6772a8100-2/)\n## [Review of Reinforcement Learning Papers #14](https://towardsdatascience.com/review-of-reinforcement-learning-papers-14-70b6772a8100-2/)\n[Artificial Intelligence](https://towardsdatascience.com/category/artificial-intelligence/)\nI present 4 publications from my research area: reinforcement learning. Let's discuss it!\n[Quentin Gallouédec](https://towardsdatascience.com/author/qgallouedec/)\nDecember 19, 20220 min read\n\n\n * [YouTube](https://www.youtube.com/c/TowardsDataScience)\n * [X](https://x.com/TDataScience)\n * [LinkedIn](https://www.linkedin.com/company/towards-data-science/)\n * [Threads](https://www.threads.net/@towardsdatascience)\n * [Bluesky](https://bsky.app/profile/towardsdatascience.com)\n\n\n[](https://towardsdatascience.com/)\nYour home for data science and AI. The world's leading publication for data science, data analytics, data engineering, machine learning, and artificial intelligence professionals.\nSubscribe to Our Newsletter[Write For TDS](https://towardsdatascience.com/submissions/)•[About](https://towardsdatascience.com/about-towards-data-science-d691af11cc2f/)•[Advertise](https://contact.towardsdatascience.com/advertise-with-towards-data-science)•[Privacy Policy](https://towardsdatascience.com/privacy-policy/)•[Terms of Use](https://towardsdatascience.com/website-terms-of-use/)\n© Insight Media Group, LLC 2026•[The New Stack](https://thenewstack.io)•[roadmap.sh](https://roadmap.sh)•[Creator Network](https://insightmediagroup.io/creator-network)\n#### Recommended Articles \nClose\n * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607686/image?etag=1782161464448&width=500&height=500) How Relevance Models Foreshadowed Transformers for NLP ](https://towardsdatascience.com/how-relevance-models-foreshadowed-transformers-for-nlp/ \"How Relevance Models Foreshadowed Transformers for NLP\")\n * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607639/image?etag=1781250125848&width=500&height=500) Javascript Fatigue: HTMX Is All You Need to Build ChatGPT — Part 1 ](https://towardsdatascience.com/javascript-fatigue-you-dont-need-js-to-build-chatgpt/ \"Javascript Fatigue: HTMX Is All You Need to Build ChatGPT — Part 1\")\n * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607930/image?etag=1782124566054&width=500&height=500) The Subset Sum Problem Solved in Linear Time for Dense Enough Inputs ](https://towardsdatascience.com/subset-sum-problem-solved-in-linear-time-for-dense-enough-inputs/ \"The Subset Sum Problem Solved in Linear Time for Dense Enough Inputs\")\n * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/3660/image?etag=1780176246070&width=500&height=500) Does Your Company Have a Data Strategy? ](https://towardsdatascience.com/does-your-company-have-a-data-strategy-e8944dbec4ae/ \"Does Your Company Have a Data Strategy?\")\n\n\n", "body_base_path": "attachments/6c6fc8bbc623/pages/", "body_markdown_path": "attachments/6c6fc8bbc623/pages/archived-body.md", "local_html": "attachments/6c6fc8bbc623/pages/archived-body.html", "body_source": "prior_capture" }