Create accountLogin

Recent searches

No recent searches

Search options

Only available when logged in.
toad.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
Mastodon server operated by David Troy, a tech pioneer and investigative journalist addressing threats to democracy. Thoughtful participation and discussion welcome.

Administered by:

Server stats:

387
active users

toad.social: AboutProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.7

#anthropic

14 posts14 participants0 posts today
News-Cafe.eu blog<p>AI poised to replace software engineers within next 12 months. Anthropic CEO predicts full automation of the coding process. <br><a href="https://www.news-cafe.eu/?go=news&amp;n=13622" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="">news-cafe.eu/?go=news&amp;n=13622</span><span class="invisible"></span></a><br><a href="https://mastodon.world/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AI</span></a> <a href="https://mastodon.world/tags/artificialintelligence" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>artificialintelligence</span></a> <a href="https://mastodon.world/tags/amodei" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>amodei</span></a> <a href="https://mastodon.world/tags/anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>anthropic</span></a> <a href="https://mastodon.world/tags/google" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>google</span></a> <a href="https://mastodon.world/tags/amazon" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>amazon</span></a> <a href="https://mastodon.world/tags/openai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>openai</span></a> <a href="https://mastodon.world/tags/technology" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>technology</span></a> <a href="https://mastodon.world/tags/meta" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>meta</span></a> <a href="https://mastodon.world/tags/zuckerberg" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>zuckerberg</span></a> @anthropic</p>
Andy Tseng<p>After <a class="hashtag" href="https://bsky.app/search?q=%23ChatGPT" rel="nofollow noopener noreferrer" target="_blank">#ChatGPT</a> and <a class="hashtag" href="https://bsky.app/search?q=%23Perplexity" rel="nofollow noopener noreferrer" target="_blank">#Perplexity</a>, <a class="hashtag" href="https://bsky.app/search?q=%23Claude" rel="nofollow noopener noreferrer" target="_blank">#Claude</a> is now stepping into the education arena. More <a class="hashtag" href="https://bsky.app/search?q=%23AI" rel="nofollow noopener noreferrer" target="_blank">#AI</a> options can mean better tools for learning, teaching, and research - if done right. Excited to see how this plays out! <a class="hashtag" href="https://bsky.app/search?q=%23Anthropic" rel="nofollow noopener noreferrer" target="_blank">#Anthropic</a> <a class="hashtag" href="https://bsky.app/search?q=%23ClaudeForEducation" rel="nofollow noopener noreferrer" target="_blank">#ClaudeForEducation</a> <a class="hashtag" href="https://bsky.app/search?q=%23Innovation" rel="nofollow noopener noreferrer" target="_blank">#Innovation</a><br><br><a href="https://www.anthropic.com/news/introducing-claude-for-education" rel="nofollow noopener noreferrer" target="_blank">Introducing Claude for educati...</a></p>
Alvin Ashcraft<p>Microsoft partners with Anthropic to create official C# SDK for Model Context Protocol. <a href="https://buff.ly/6lQxN52" rel="nofollow noopener noreferrer" target="_blank">buff.ly/6lQxN52</a> <a class="hashtag" href="https://bsky.app/search?q=%23ai" rel="nofollow noopener noreferrer" target="_blank">#ai</a> <a class="hashtag" href="https://bsky.app/search?q=%23csharp" rel="nofollow noopener noreferrer" target="_blank">#csharp</a> <a class="hashtag" href="https://bsky.app/search?q=%23dotnet" rel="nofollow noopener noreferrer" target="_blank">#dotnet</a> <a class="hashtag" href="https://bsky.app/search?q=%23mcp" rel="nofollow noopener noreferrer" target="_blank">#mcp</a> <a class="hashtag" href="https://bsky.app/search?q=%23anthropic" rel="nofollow noopener noreferrer" target="_blank">#anthropic</a> <a class="hashtag" href="https://bsky.app/search?q=%23oss" rel="nofollow noopener noreferrer" target="_blank">#oss</a><br><br><a href="https://buff.ly/6lQxN52" rel="nofollow noopener noreferrer" target="_blank">Microsoft partners with Anthro...</a></p>
Alvin Ashcraft 🐿️<p>Microsoft partners with Anthropic to create official C# SDK for Model Context Protocol.</p><p><a href="https://devblogs.microsoft.com/blog/microsoft-partners-with-anthropic-to-create-official-c-sdk-for-model-context-protocol" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">devblogs.microsoft.com/blog/mi</span><span class="invisible">crosoft-partners-with-anthropic-to-create-official-c-sdk-for-model-context-protocol</span></a></p><p><a href="https://hachyderm.io/tags/ai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ai</span></a> <a href="https://hachyderm.io/tags/csharp" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>csharp</span></a> <a href="https://hachyderm.io/tags/dotnet" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>dotnet</span></a> <a href="https://hachyderm.io/tags/mcp" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>mcp</span></a> <a href="https://hachyderm.io/tags/anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>anthropic</span></a> <a href="https://hachyderm.io/tags/oss" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>oss</span></a></p>
Victoria Stuart 🇨🇦 🏳️‍⚧️<p>Researchers lift the lid on how reasoning models actually “think”<br><a href="https://www.economist.com/science-and-technology/2025/04/02/researchers-lift-the-lid-on-how-reasoning-models-actually-think" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">economist.com/science-and-tech</span><span class="invisible">nology/2025/04/02/researchers-lift-the-lid-on-how-reasoning-models-actually-think</span></a><br>nonpaywalled: <a href="https://archive.fo/pn6du" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="">archive.fo/pn6du</span><span class="invisible"></span></a></p><p>Tracing the thoughts of a large language model<br><a href="https://www.anthropic.com/research/tracing-thoughts-language-model" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">anthropic.com/research/tracing</span><span class="invisible">-thoughts-language-model</span></a><br><a href="https://news.ycombinator.com/item?id=43495617" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">news.ycombinator.com/item?id=4</span><span class="invisible">3495617</span></a></p><p><a href="https://mastodon.social/tags/LLM" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>LLM</span></a> <a href="https://mastodon.social/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> <a href="https://mastodon.social/tags/Claude" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Claude</span></a> <a href="https://mastodon.social/tags/reasoning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>reasoning</span></a> <a href="https://mastodon.social/tags/ChainOfThought" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ChainOfThought</span></a></p>
Jason Yip<p>What is <a href="https://mastodon.online/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AI</span></a> thinking? <a href="https://mastodon.online/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> researchers are starting to figure it out <a href="https://www.fastcompany.com/91309820/what-is-ai-thinking-anthropic-researchers-are-starting-to-figure-it-out" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">fastcompany.com/91309820/what-</span><span class="invisible">is-ai-thinking-anthropic-researchers-are-starting-to-figure-it-out</span></a></p>
ResearchBuzz: Firehose<p>Engadget: Claude’s new Learning mode will prompt students to answer questions on their own . “At the heart of Claude for Education is a new Learning mode that changes how Anthropic’s chatbot interacts with users. With the feature engaged, Claude will attempt to guide students to a solution, rather than providing an answer outright, when asked a question. It will also employ the Socratic […]</p><p><a href="https://rbfirehose.com/2025/04/03/engadget-claudes-new-learning-mode-will-prompt-students-to-answer-questions-on-their-own/" class="" rel="nofollow noopener noreferrer" target="_blank">https://rbfirehose.com/2025/04/03/engadget-claudes-new-learning-mode-will-prompt-students-to-answer-questions-on-their-own/</a></p>
Kris Shrishak<p>Last year <a href="https://eupolicy.social/tags/European" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>European</span></a> Parliament Archive unit head appeared in promo videos for Anthropic when EP began using <a href="https://eupolicy.social/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a>'s <a href="https://eupolicy.social/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AI</span></a> model.</p><p>ICCL Enforce investigated.</p><p>There are no contracts. No impact assessment. No bias checks. Blind faith in Constitutional AI and <a href="https://eupolicy.social/tags/Amazon" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Amazon</span></a> lock-in.</p><p><a href="https://www.iccl.ie/press-release/how-not-to-deploy-generative-ai-the-story-of-the-european-parliament/" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">iccl.ie/press-release/how-not-</span><span class="invisible">to-deploy-generative-ai-the-story-of-the-european-parliament/</span></a></p>
PKPs Powerfromspace1<p>@mattberman on YT 📺 </p><p><a href="https://mstdn.social/tags/anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>anthropic</span></a> <a href="https://mstdn.social/tags/llm" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>llm</span></a> <a href="https://mstdn.social/tags/ai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ai</span></a> <a href="https://mstdn.social/tags/safety" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>safety</span></a></p><p>We Finally Figured Out How AI Actually Works… (not what we thought!)" </p><p><a href="https://youtu.be/4xAiviw1X8M?feature=shared" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">youtu.be/4xAiviw1X8M?feature=s</span><span class="invisible">hared</span></a></p>
🔘 G◍M◍◍T 🔘<p>💡 MCP: lo standard che collega modelli AI e dati in tempo reale</p><p><a href="https://gomoot.com/mcp-lo-standard-che-collega-modelli-ai-e-dati-in-tempo-reale/" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">gomoot.com/mcp-lo-standard-che</span><span class="invisible">-collega-modelli-ai-e-dati-in-tempo-reale/</span></a></p><p><a href="https://mastodon.uno/tags/anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>anthropic</span></a> <a href="https://mastodon.uno/tags/blog" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>blog</span></a> <a href="https://mastodon.uno/tags/database" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>database</span></a> <a href="https://mastodon.uno/tags/dati" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>dati</span></a> <a href="https://mastodon.uno/tags/llm" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>llm</span></a> <a href="https://mastodon.uno/tags/mcp" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>mcp</span></a> <a href="https://mastodon.uno/tags/news" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>news</span></a> <a href="https://mastodon.uno/tags/openai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>openai</span></a> <a href="https://mastodon.uno/tags/picks" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>picks</span></a> <a href="https://mastodon.uno/tags/server" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>server</span></a> <a href="https://mastodon.uno/tags/tech" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>tech</span></a> <a href="https://mastodon.uno/tags/tecnologia" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>tecnologia</span></a></p>
Vladimir Savić<p>Tracing the thoughts of a large language model <a href="https://www.anthropic.com/news/tracing-thoughts-language-model" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">anthropic.com/news/tracing-tho</span><span class="invisible">ughts-language-model</span></a> <a href="https://mastodon.social/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AI</span></a> <a href="https://mastodon.social/tags/GenAI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>GenAI</span></a> <a href="https://mastodon.social/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> <a href="https://mastodon.social/tags/LLM" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>LLM</span></a></p>
IT News<p>MCP: The new “USB-C for AI” that’s bringing fierce rivals together - What does it take to get OpenAI and Anthropic—two competitors in the AI as... - <a href="https://arstechnica.com/information-technology/2025/04/mcp-the-new-usb-c-for-ai-thats-bringing-fierce-rivals-together/" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">arstechnica.com/information-te</span><span class="invisible">chnology/2025/04/mcp-the-new-usb-c-for-ai-thats-bringing-fierce-rivals-together/</span></a> <a href="https://schleuss.online/tags/modelcontextprotocol" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>modelcontextprotocol</span></a> <a href="https://schleuss.online/tags/largelanguagemodels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>largelanguagemodels</span></a> <a href="https://schleuss.online/tags/machinelearning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>machinelearning</span></a> <a href="https://schleuss.online/tags/anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>anthropic</span></a> <a href="https://schleuss.online/tags/chatgpt" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>chatgpt</span></a> <a href="https://schleuss.online/tags/chatgtp" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>chatgtp</span></a> <a href="https://schleuss.online/tags/biz" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>biz</span></a>⁢ <a href="https://schleuss.online/tags/openai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>openai</span></a> <a href="https://schleuss.online/tags/api" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>api</span></a> <a href="https://schleuss.online/tags/mcp" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>mcp</span></a> <a href="https://schleuss.online/tags/ai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ai</span></a></p>
Wulfy<p>No, <a href="https://infosec.exchange/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AI</span></a> frontier models don't "just guess words", it's far more complicated than that.</p><p><a href="https://infosec.exchange/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> built an <a href="https://infosec.exchange/tags/LLM" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>LLM</span></a> "brain scanner" (so far AIs have been black boxes).</p><p>According to Anthropic, "it currently takes a few hours of human effort to understand the circuits we see, even on prompts with only tens of words." And the research doesn't explain how the structures inside LLMs are formed in the first place.</p>
PKPs Powerfromspace1<p><span class="h-card" translate="no"><a href="https://schleuss.online/@itnewsbot" class="u-url mention" rel="nofollow noopener noreferrer" target="_blank">@<span>itnewsbot</span></a></span> now... they are just doing it now 🙄 <a href="https://mstdn.social/tags/ai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ai</span></a> <a href="https://mstdn.social/tags/anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>anthropic</span></a></p>
Rod2ik 🇪🇺 🇨🇵 🇪🇸 🇺🇦 🇨🇦 🇩🇰 🇬🇱<p><a href="https://mastodon.social/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> vient de publier des études révélant comment son modèle <a href="https://mastodon.social/tags/Claude" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Claude</span></a> "réfléchit" réellement. </p><p>Les chercheurs ont découvert que l' <a href="https://mastodon.social/tags/IA" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>IA</span></a> <a href="https://mastodon.social/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AI</span></a> planifie ses réponses à l'avance, pense dans un <a href="https://mastodon.social/tags/langage" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>langage</span></a> conceptuel <a href="https://mastodon.social/tags/universel" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>universel</span></a> et peut même parfois fournir des explications qui ne reflètent pas son véritable processus interne.</p><p><a href="https://www.lesnumeriques.com/intelligence-artificielle/comment-fonctionne-vraiment-une-ia-les-chercheurs-d-anthropic-ont-enfin-un-debut-de-reponse-n234978.html" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">lesnumeriques.com/intelligence</span><span class="invisible">-artificielle/comment-fonctionne-vraiment-une-ia-les-chercheurs-d-anthropic-ont-enfin-un-debut-de-reponse-n234978.html</span></a></p>
Victoria Stuart 🇨🇦 🏳️‍⚧️<p>On the Biology of a Large Language Model [Claude LLM]<br><a href="https://transformer-circuits.pub/2025/attribution-graphs/biology.html" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">transformer-circuits.pub/2025/</span><span class="invisible">attribution-graphs/biology.html</span></a><br>Anthropic: On the Biology of a Large Language Model : MachineLearning<br><a href="https://old.reddit.com/r/MachineLearning/comments/1jmhoq6/r_anthropic_on_the_biology_of_a_large_language" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">old.reddit.com/r/MachineLearni</span><span class="invisible">ng/comments/1jmhoq6/r_anthropic_on_the_biology_of_a_large_language</span></a></p><p><a href="https://mastodon.social/tags/LLM" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>LLM</span></a> <a href="https://mastodon.social/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> <a href="https://mastodon.social/tags/Claude" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Claude</span></a> <a href="https://mastodon.social/tags/ChainOfThought" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ChainOfThought</span></a> <a href="https://mastodon.social/tags/reasoning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>reasoning</span></a></p>
Victoria Stuart 🇨🇦 🏳️‍⚧️<p>/1 That post includes the following video - which in simple language / examples a basic overview of a current, reasoning LLM (thought; Claude) and "prompt engineering."</p><p>Tracing the thoughts of a large language model<br><a href="https://www.youtube.com/watch?v=Bj9BD2D3DzA" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="">youtube.com/watch?v=Bj9BD2D3DzA</span><span class="invisible"></span></a></p><p><a href="https://mastodon.social/tags/LLM" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>LLM</span></a> <a href="https://mastodon.social/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> <a href="https://mastodon.social/tags/Claude" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Claude</span></a> <a href="https://mastodon.social/tags/ChainOfThought" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ChainOfThought</span></a> <a href="https://mastodon.social/tags/reasoning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>reasoning</span></a></p>
Miguel Afonso Caetano<p>"Why do language models sometimes hallucinate—that is, make up information? At a basic level, language model training incentivizes hallucination: models are always supposed to give a guess for the next word. Viewed this way, the major challenge is how to get models to not hallucinate. Models like Claude have relatively successful (though imperfect) anti-hallucination training; they will often refuse to answer a question if they don’t know the answer, rather than speculate. We wanted to understand how this works.</p><p>It turns out that, in Claude, refusal to answer is the default behavior: we find a circuit that is "on" by default and that causes the model to state that it has insufficient information to answer any given question. However, when the model is asked about something it knows well—say, the basketball player Michael Jordan—a competing feature representing "known entities" activates and inhibits this default circuit (see also this recent paper for related findings). This allows Claude to answer the question when it knows the answer. In contrast, when asked about an unknown entity ("Michael Batkin"), it declines to answer.</p><p>Sometimes, this sort of “misfire” of the “known answer” circuit happens naturally, without us intervening, resulting in a hallucination. In our paper, we show that such misfires can occur when Claude recognizes a name but doesn't know anything else about that person. In cases like this, the “known entity” feature might still activate, and then suppress the default "don't know" feature—in this case incorrectly. Once the model has decided that it needs to answer the question, it proceeds to confabulate: to generate a plausible—but unfortunately untrue—response."</p><p><a href="https://www.anthropic.com/research/tracing-thoughts-language-model" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">anthropic.com/research/tracing</span><span class="invisible">-thoughts-language-model</span></a></p><p><a href="https://tldr.nettime.org/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AI</span></a> <a href="https://tldr.nettime.org/tags/GenerativeAI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>GenerativeAI</span></a> <a href="https://tldr.nettime.org/tags/LLMs" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>LLMs</span></a> <a href="https://tldr.nettime.org/tags/Chatbots" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Chatbots</span></a> <a href="https://tldr.nettime.org/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> <a href="https://tldr.nettime.org/tags/Claude" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Claude</span></a> <a href="https://tldr.nettime.org/tags/Hallucinations" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Hallucinations</span></a></p>
Victoria Stuart 🇨🇦 🏳️‍⚧️<p>Tracing the thoughts of a large language model<br><a href="https://www.anthropic.com/research/tracing-thoughts-language-model" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">anthropic.com/research/tracing</span><span class="invisible">-thoughts-language-model</span></a><br><a href="https://news.ycombinator.com/item?id=43495617" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">news.ycombinator.com/item?id=4</span><span class="invisible">3495617</span></a></p><p><a href="https://mastodon.social/tags/LLM" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>LLM</span></a> <a href="https://mastodon.social/tags/Anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Anthropic</span></a> <a href="https://mastodon.social/tags/Claude" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>Claude</span></a> <a href="https://mastodon.social/tags/reasoning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>reasoning</span></a> <a href="https://mastodon.social/tags/ChainOfThought" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ChainOfThought</span></a></p>
PKPs Powerfromspace1<p>@wired.com</p><p>BY STEVEN LEVY<br>BUSINESS<br>MAY 21, 2024 11:00 AM<br>AI Is a Black Box. Anthropic Figured Out a Way to Look Inside</p><p><a href="https://www.wired.com/story/anthropic-black-box-ai-research-neurons-features/" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">wired.com/story/anthropic-blac</span><span class="invisible">k-box-ai-research-neurons-features/</span></a></p><p><a href="https://mstdn.social/tags/ai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>ai</span></a> <a href="https://mstdn.social/tags/llm" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>llm</span></a><br><a href="https://mstdn.social/tags/safety" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>safety</span></a> <a href="https://mstdn.social/tags/alignment" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>alignment</span></a> <a href="https://mstdn.social/tags/anthropic" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>anthropic</span></a></p>
ExploreLive feeds
About