There's definitely something here. Big risk though is that we end up re-building integrations between LLMs and other software because we need to "feed the inbox". Resist the urge! Check out `mcp-subscribe`: x.com/syllogistic/st…

Dear Santa, If we get some open-source reasoning models this week, all I wish for is tool-calling inside the chain of thought. P.S. My agent has been a good boy. cc AI at Meta DeepSeek

Naturally "found an antidote" == "changed the system prompt"

Hot take: the llm sdk is the mcp client

Made some notes on how Cursor works under the hood based on their security documentation - it turns out an organization's list of subprocessors offers a loose form of "view source" for their infrastructure! simonwillison.net/2025/May/11/cu…

What exactly happens when neo learns kung fu?

So from the screenshots / videos online the grok companion is definitely a 3d model audio 2 face thing, not a diffusion / veo3 style model. Grok

gpt-oss is a big deal; it is a state-of-the-art open-weights reasoning model, with strong real-world performance comparable to o4-mini, that you can run locally on your own computer (or phone with the smaller size). We believe this is the best and most usable open model in the

This is the new unit testing

I have three kids under 6 and just want to take a minute to appreciate magnatiles.

Circa 2011 I went to FLAVOR FLAV's hot chicken restaurant in Detroit. My buddy asked if he could get some extra sauce. The dude at the register asked the manager who replied, what would Falv do? We got the extra sauce. They went out of business a few months later.

openai codex wacky inference bug? Alexander Embiricos

Resources, the most underused thing in MCP, about to blow up

Let me pitch you Skills: - They allow anyone to customize agents with a simple primitive — files - They give us continuous learning until continuous learning arrives - They are powerful, composable, and sufficiently AGI-pilled - you can now say “probably a skill issue” at work

Yes this is simply a public service announcement that you should give your file system enabled agents markdown files with headers for lazy loading. Also because it comes from Anthropic it will actually stick as a standard. Skills are both (annoyingly) obvious in hindsight and

nanochat base model: 68 hours on a single 5090 made my basement feel nice and toasty