Jonathan Chang

inference @poolsideai, building

[Jonathan Chang @ChangJonathanC]

while we wait for gpt-5 to drop. Here is a flex attention tutorial for building a < 1000 LoC vllm from scratch
https://t.co/PVyauMezM3

[Jonathan Chang @ChangJonathanC]

@vikhyatk thanks! wanna read a early version when it's ready?

[Jonathan Chang @ChangJonathanC]

I'm very excited to write this next blog post

[Jonathan Chang @ChangJonathanC]

@kimbochen you can create a thread of threads

[Jonathan Chang @ChangJonathanC]

ARC-AGI robotics

[DiscussingFilm @DiscussingFilm]

First look at the LEGO Sagrada Família

It is the largest LEGO set ever made at 12,060 pieces.

Will cost $800.

[kalomaze @kalomaze]

lol tried this with a logit lens

[Jonathan Chang @ChangJonathanC]

I don't buy almond from supermarkets anymore

[Jonathan Chang @ChangJonathanC]

🤔

[Jonathan Chang @ChangJonathanC]

@willdepue Why not just use hook?

[Jonathan Chang @ChangJonathanC]

@sharifshameem

[Jonathan Chang @ChangJonathanC]

accidentally discovered a way to get raw-ish cot from gpt-5.5

here is it trying to solve IMO 2025 P1

[Jonathan Chang @ChangJonathanC]

codex psychosis

[OpenAI @OpenAI]

It's time to fly.

[Jonathan Chang @ChangJonathanC]

fork is now in Claude Code
https://t.co/HKzSxWQl95

[ClaudeDevs @ClaudeDevs]

We've updated /fork in Claude Code

/fork now runs a background agent with your exact context (system prompt, tools, history, model) and prompt cache. The result gets returned to your session.

/branch (the old /fork) still copies the transcript to a new session you drive.

[Jonathan Chang @ChangJonathanC]

Claude Code's dispatch_agent is cool, it's like spawning a new process, but what if you can call fork()?

I think it'll be a lot more powerful, so I created a repo to explore this pattern 👇

[Jonathan Chang @ChangJonathanC]

github repo: https://t.co/dF5zaSUyTW

[Jonathan Chang @ChangJonathanC]

@tszzl people have to do faster model so as not Amdahls law the autoresearch agent running on 8xB200 gpus

[Jonathan Chang @ChangJonathanC]

@benhylak @justinsunyt didn't OpenAI use it for scaling image-1 in chatgpt?

[AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes]

It's up to you to break generational trauma

[Jonathan Chang @ChangJonathanC]

@stochasticchasm @mermachine bpe with byte fallback can explain it

[Jonathan Chang @ChangJonathanC]

Can't believe Microsoft released this

[Jonathan Chang @ChangJonathanC]

@thsottiaux Sponsored post

[Jonathan Chang @ChangJonathanC]

Yes the AI is writing unnecessarily defensive code, but are you doing unnecessarily defensive code review?

[Jonathan Chang @ChangJonathanC]

@drakefjustin @VitalikButerin @pangram slop?