66 mentions · 68 chunks · 43 episodes
...t injection. Bruce Schneier: Why AI Keeps Falling for Prompt Injection Attacks. Anthropic quietly fixed flaws in its Git MCP server that allowed for remote code execution. Red Teaming BrowseSafe: Prompt Injection Risks in Perplexity's Open...
...antified. Modern society acts like "If it can't be measured it doesn't matter." Anthropic's Economic Report falls into the same gap. You play the board game not to win the board game but for the fun of the struggle of playing the board gam...
...ociety acts like most things are modular, but actually most things are complex. Anthropic's Economic Report falls into this trap.
Anthropic's Red Team's Agents, in simulated testing, found $4.6M in Smart Contracts exploits.
...rsuasion Costs LLMs can be skewed by just a few hundred aligned inputs (see the Anthropic paper). The same is true for society. A small number of strongly aligned components can have an outsize impact on the whole. See also The Ouija Board...
...ing the "I believe LLMs may have functional emotions in some sense": "I believe Anthropic may have functional emotions in some sense. Not necessarily identical to human emotions, but analogous processes." Thought provoking! We have a numbe...
...ot be great if most LLM usage in the US is an open-source Chinese model. First, Anthropic's research shows it's remarkably easy to poison a model of arbitrary size with deliberately chosen malicious training data. Second, if there's a mode...
...omises. Not perfect, but radically better. This should be table stakes! OpenAI, Anthropic, every frontier lab: make this the new baseline. This is like what Apple did with Private Cloud Compute, but with a frontier model. Anthropic does us...
Anthropic caught real agent-swarms attacks being orchestrated in Claude. Remember: these tools help the baddies, too!
Anthropic at some point had a billboard ad campaign: "You've got a friend in Claude." Even Anthropic, who doesn't have a viable consumer strategy, is falling i...
...ther hosts, like Go did for packages. And b) the fact it's two distinct actors (Anthropic and Github) makes it much less likely there will be bad behavior than if those two positions were merged into one.
...e raising capital when others can't. But a stable equilibrium to me seems to be Anthropic and Google staying in the race indefinitely. They have the capital and backing to stay in the game no matter what. And of course, there's the possibi...
...gnitude more powerful techniques for getting more value out of it. For example, Anthropic Skills, which Simon thinks is a huge deal. MCP felt cool, but it had a low ceiling and was easy to overwhelm the context window. Really what matters ...
Anthropic Skills is powerful for the same reason my old Code Sprouts project felt unreasonably powerful to me a few years ago. Giving just a teensy bit of stru...
Anthropic keeps doing something simple and elegant… That also causes traffic jams. Impeccable first-order thinking. Non-existent second-order thinking.[iu]
Anthropic showed how a very small number of samples in training can poison even large model's outputs. This isn't a surprise to me. Even a small bias, if it's ...
Anthropic is about to release a feature of LLM-powered software. Opal from Google is a similar shape, but without custom UI. Normal UI, but LLM guts underneath...
Anthropic is the model company whose incentives I trust the most. That's because they don't have a viable consumer play. It's the consumer plays that push towa...
A new important concept: "vibehacking[ai]" / "vibepwned" Anthropic: "Agentic AI has been weaponized. AI models are now being used to perform sophisticated cyberattacks, not just advise on how to carry them out." Reme...
Anthropic changed their policy to train on messages in the Claude consumer experience. This is a small signal they don't believe AGI is right around the corner...