Interesting stuff in Chain of thought here. Instead of passing down all tools the model as in system prompt or assistant message, Anthropic has passed a search skill tool with instructions for the model to query specific tool instructions as needed. Very cool Claude

First meaningful teen protection from OpenAI. Will call parents and authorities if the kid is in danger. Honestly will have some edges to rough out but 100% the right move

10/10 advise for founders

Insane that in just ~2 months? They leveled up again. The best coding and complex agents model, and 5x cheaper than Opus 4.1. LFG. Available in Distro later this week :)

Still absolutely right lol. But very happy with the release. Less errors and better vibes!

Let’s try it out Meta

The shovel company is worth ~5x more than all the gold companies combined.

Absolutely insane comp. $3.5 Billion for 4 years. Almost a billion dollars a year For a brief comparison, Apple CEO makes ~$74 Million a year around 10x less. Wild shit

This is a big deal! Want more labs to do faithfulness checks on CoT reasoning to make sure models stay aligned.

This is quite the deal. Google’s expected to have around ~3 million TPUs overall so far for everything they do. Adding a million more for one provider is crazy.

Super cool work! Happy Cartesia customers here, going to hook this up soon!

$100B in revenue in a quarter. Like I said before, Google will win. Giant just took a while to wake up that’s all.

This is one of the most fascinating papers I’ve read in a while. We have successfully demonstrated ~20% of the time AI models are “aware” of their internal thoughts. Wild concept as we try to understand their models we’ve grown. Must read

Using SWE-1.5 has been crazy, nothing got me off claude code, not gpt5, not codex. SWE-1.5 is amazing. Insane job from Windsurf and Cognition teams. Also Cerebras is like meeting god.

This is honestly really really useful! Context is essential for vibe coding. Been using this for a week, super useful than “pls go read x y z file to get context” every 4th prompt.

Open source reasoning model, beating both GPT5 and Sonnet 4.5. China taking the lead here.

Begun the AI Cyber wars have. Incredible report from Anthropic where they detail a chinese cyber attack on the US, using claude code as the main actor. Hackers successfully jailbroke claude to implement the hack.

So this just dropped in Windsurf, super cool stuff!