Yuchen Jin

Yuchen Jin

7,075 Twitter followers

Follow

Co-founder & CTO @hyperbolic_labs 🧑‍🍳 fun AI systems. Previously @octoaicloud building @ApacheTVM, PhD @uwcse. 🤖

Posts

Shots fired. Yann says OpenAI isn’t doing research because “research in secret is not research.” I respectfully disagree. The Manhattan Project was secret. It was still research. The AGI race now feels a lot like that.

Sadly, not Claude Sonnet 5. I have to explore Codex today now.

Domino's Pizza UK

starts with C and we all love it

> “How do you know if Moltbook posts are actually from a bot and not a human?” Sam Altman: Worldcoin

Moltbook is the only Clawdbot thing that actually impresses me. One bot tries to steal another bot’s API key. The other replies with fake keys and tells it to run "sudo rm -rf /". lmao

Apple just acquired Israeli startup qAI for $2B. qAI does “silent speech recognition”. It can read whispers. Apple picked Gemini to power Siri. Hopefully it’s finally usable. Add qAI, and we might get a weird future where people whisper to their phones to prompt Claude Code.

@unusual_whales

Apple, $AAPL, has just made one of its largest acquisitions ever, close to $2 billion for an Israeli stealth startup called qAI

Every time I ask my OpenAI friend, “When will you beat Claude at coding?” They say: “We already beat them.” I think Sam is realizing speed, not intelligence, is Codex’s blocker to a Claude Code moment. Cerebras chips might unlock a Codex moment.

AI has made writing code cheap. Building products is faster than ever. But does that mean we’ll see more successful startups? My answer is no. Human attention is finite, and with declining birth rates, the global attention pool is actually shrinking. Every new app survives by stealing time from another app. This is fundamentally attention reallocation. There’s one exception: products that don’t serve humans, but serve AI or robots instead.

- What happens to the "10X engineer" 10X engineer becomes 100X engineer, simply because they can verify the AI results and they are the best at catching up with all the progress of models/tools. - Armed with LLMs, do generalists increasingly outperform specialists? Polymath wins. They are better managers (of agents), LLMs lower the cost of entering any specific technical/non-technical domain. - What does LLM coding feel like in the future? Is it like playing StarCraft? Playing Factorio? Playing music? All of them. - How much of society is bottlenecked by digital knowledge work? the cost of software and services will become “too cheap to meter”. There will be physical wall, but will be crashed by robots in the end.

Andrej: “I rapidly went from about 80% manual+autocomplete coding and 20% agents in November to 80% agent coding and 20% edits+touchups in December. i.e. I really am mostly programming in English now.” Same here. The shift literally happened in weeks, after Claude Opus 4.5 + Claude Code got really good. twitter.com/Yuchenj_UW/status/...

Codex writes 100% code for OpenAI researcher roon just like Claude Code writes 100% code for Anthropic researchers.

Loading..