66 mentions · 68 chunks · 43 episodes
...turally addressing prompt injection, LLM agents can't safely reach mass market. Anthropic's 11% attack success Simon Willison calls a "catastrophic failure rate" "Smarter models" hit asymptotic returns. A structural approach is necessary t...
Anthropic announced Claude for Chrome this week. Their blog post announcing it mentioned it will be available to a small set of users because they haven't yet ...
Anthropic rolled out a new safety feature optimized for "model welfare" Obviously this is a reasonable feature given the topics that it cuts off. But the frame...
Anthropic released a deeper paper on the agentic misalignment. That is, how the model would choose to blackmail its creators in some cases. Simon Willison's su...
With OpenAI and Anthropic the model is the product. The product is a model in the middle like a christmas tree decorated with various doo dads. The doo dads are useful, but th...
Another thought on emergence from Anthropic's guide to multi-agent systems: "Once intelligence reaches a threshold, multi-agent systems become a vital way to scale performance. For instance, al...
... likely reserve their model for their own 1P product. Other leading models from Anthropic and Google likely would have done the same. But luckily we live in the world where OpenAI had already released their API before ChatGPT got big. Beca...
ChatGPT[mk] will tell you how it would deceive you if you ask it. Anthropic will say "There must be some mistake, I would never deceive you."[ml] Which do you believe?
Anthropic's research on the inner workings of LLMs is fascinating. They're studying LLMs less like an engineer would study a technical artifact and more like a...
... for a model to over-fit to a specific framework, like React. If I'm right that Anthropic has specially focused on React, I'd imagine the model got at least incrementally worse on non-React code. The model likely "pulls" more heavily towar...
...al difference between 100% accuracy and 99% accuracy." "I feel like [OpenAI and Anthropic] have gone to market ahead of product-market fit. I feel like the prompt looks like a product but isn't, or it's only a product for certain segments,...
...drive on, as long as everyone in the country picks the same side. It looks like Anthropic will be making a registry akin to npm. This totally makes sense for them to do! Being that schelling point in the ecosystem of maintaining the most c...
... is much better. I love this feature, but it seems like a strategic misstep for Anthropic. Typically you want your product to be sticky, and one way to do that is to encourage users to store state that makes their experience better and bet...
... reasons people are implicitly combining them in their heads is because OpenAI, Anthropic, and Google all have entrants in both levels. But this is more an artifact of the "vertical integration for proof of existence" phase of the new para...
Anthropic Artifacts is 100% frog DNA. It can whip up a little interactive thing for you based on an English language prompt. But all it has to work with is wha...
...le these kinds of interactions together by starting new conversations, or using Anthropic's projects, or manually cobbling together tools on top of the raw API. It feels like I'm banging my head against a command line interface, wishing fo...
...of prose at different layers of distillation for ~free seems like a big unlock! Anthropic's practical guide to agents is excellent. Grounded, clarifying, direct, insightful. Scaling Test Time Compute from HuggingFace. Great overview of the...
...be "dangerous" …and also undermine their app's differentiation and power. Later Anthropic comes along and does the same. Neither feels compelled to release an API because both want to be an aggregator. In that world, we'd have LLM-powered ...
...for what plain old code can do. That leads to, for example, trying to create an Anthropic Artifact to identify what kind of dog is in the picture. But Artifacts can't call out to LLMs when they are executed. They are compiled from english ...
...e a low-hanging fruit made possible by LLMs, just waiting to be discovered. How Anthropic Built Artifacts: https://newsletter.pragmaticengineer.com/p/how-anthropic-built-artifacts The feature went from initial demo to production launch in ...