Andy Xor

Actual AI manifesto: "Actual" means "not fake". We believe 'deep learning' is a dead-end and actual AI will come from understanding animal cognition (including human neuroscience). One has to respect the constraints of biological learning & memory in order to imitate them.

It's wild that every time you run a Codex code review from Claude Code, it finds critical issues. Not 95% of the times, 100%.

Holy shit… someone just made Claude instances talk to each other. Not APIs. Not agents. Not orchestrators. Just multiple Claude Code sessions… messaging each other like coworkers. It’s called claude-peers — and it turns one Claude into a team. Here’s what’s happening: Run 5 Claude Code sessions across different projects Each one auto-discovers the others They send messages instantly Ask questions Share context Coordinate work Your AI tools literally collaborate. Example: Claude A (poker-engine): "what files are you editing?" Claude B (frontend): "working on auth.ts + UI state" Claude A: "ok I'll avoid touching auth logic" No conflicts. No manual coordination. Just AI syncing itself. Under the hood: • Local broker daemon (localhost) • SQLite peer registry • MCP servers per session • Instant channel push messaging • Auto peer discovery • Cross-project communication Everything runs locally. No cloud. No latency. What it unlocks: • Multi-agent coding without frameworks • One Claude writes backend, another frontend • One debugs while another refactors • Research Claude feeds builder Claude • Large projects split across AI workers This is basically: "spawn 5 Claudes and let them coordinate themselves" Even crazier: Each instance auto-summarizes what it's doing Other Claudes can see: • working directory • git repo • current task • active files They know what the others are working on. Commands: • list_peers → find all Claude sessions • send_message → talk to another Claude • set_summary → describe your task • check_messages → manual fallback So you can literally say: "message peer 3: what are you working on?" …and it responds instantly. No orchestration layer. No agent framework. Just Claudes… talking. This is the cleanest multi-agent system I've seen. We're moving from: 1 AI assistant → to AI teams that coordinate themselves. And it's all running on your machine. Wild.

OpenAI's latest repo has Claude as the third top contributor 😭😂

SF public schools finally ends its 12-year experiment in removing the option of taking algebra in the eighth grade in pursuit of racial equity. Every retarded left-wing nostrum that has no possibility of working gets a decade+ of runway before we have enough studies to prove that the thing that could never work didn't work. The key problem of public policy must be how to shut down this pipeline, which begins by mass defunding of corrupted academic programs and NGO's.

Eric Schmidt says the 10x advantage is no longer execution. It is defining what counts as success. A programmer writes a spec and an evaluation function, runs it at 7pm, and wakes up to what was invented overnight. The advantage now belongs to whoever can specify the problem precisely. The rest will be automated.

One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.

Little known fact, the Anthropic Labs team (the team I joined Anthropic to be on) shipped: - MCP - Skills - Claude Desktop app - Claude Code It was just a few of us, shipping fast, trying to keep pace with what the model was capable of. Those early Desktop computer use prototypes, back in the Sonnet 3.6 days, felt clunky and slow. But it was easy to squint and imagine all the ways people might use it once it got really good. Fast forward to today. I am so excited to release full computer use in Cowork and Dispatch. Really excited to see what you do with it!

BREAKING: Wikipedia has banned editor after investigation reveals he was responsible for over a million pro-Hamas edits on the site.

This is incredibly generous. TSA agents across the country are relying on food pantries and community donations just to get by. I remain the lone Dem to vote with my Republican colleagues to fully fund DHS and get people paid. It should never come to this point.

Your honor I told Claude not to make any mistakes

Reddit mods are running damage control for Mamdani's wife

Andrej Karpathy says when AI agents fail, it's usually a skill issue, not a capability issue You didn't write good enough instructions, didn't set up the right memory tool, or didn't parallelize correctly "the real shift is working in macro actions" One does research, one writes code, one plans, all running 20-minute tasks simultaneously

“Kids I have good news. Daddy is out of Claude tokens until 3PM. He has time to play with you now.”

I have so much gratitude to people who wrote extremely complex software character-by-character. It already feels difficult to remember how much effort it really took. Thank you for getting us to this point.

Me reviewing the code written by Claude before pushing it to production

AI career: 1. Get PhD from top school 2. Work for DeepMind/OpenAI/Anthropic for 6 months 3. Quit and start your own neolab with billions in funding 4. Fail to achieve AGI but retire rich

Somehow, everybody is a victim, except for the Jewish community that was attacked.

The New York Times is absolutely beyond saving. Insanity.

The Left is so committed to the position that bad guys are actually good guys and good guys are actually bad guys that multiple major media outlets tried to justify and rationalize the actions of a terrorist who drove a truck full of explosives into a pre-school full of babies.

"Why did you identify this as one of the craziest grants?" "Because it references feminist and queer insights into prison abolition and LGBTQ studies." "Any other reasons?" "No." Legendarily based. 😆