Recording my Tree of Thoughts here.
In Rust we trust.
All Distro-hoppers lead to NixOS.
My WiFi hotspot name is Hyperliquid and the password is jeffyan@
Taiwan, Japan or Korea could have saved Jensen -- by sending domestic baddie spies to keep him occupied. Not random baddies but those with a real career in the industry. A lot of them would probably volunteer to do it.
Still spying and still cheating, but at least not connected to Beijing.
https://t.co/H9RQZDX113
A Reddit comment of the Whip app.
I build it to free myself from being stuck with my laptop, and it has become the absolute best product for people like me.
Since it is open software I did not really profit from it.
Anyway, comments like this and new GitHub ⭐ make my day😄.
https://t.co/KTr5tHLJ51
I asked Codex to implemented based on the paper anyway.
The paper claimed that with LLM-generated weight masks the recognition accuracy is +3% higher than the backbone while my result showed -2% degradation.
I tried to reproduce a CVPR 2026 paper whose code is on GitHub.
Codex told me the repo is missing training modules and data.
They claimed to open source it but left the core parts out.
An issue was opened by someone in June but got ignored.
How did a reviewer judge a CV paper? They just scan it and run no code?
If you check systems papers from OSDI/SOSP, reproducibility is very important. They usually open source the full codebase with a detailed guide for reproduction.
I tried to reproduce a CVPR 2026 paper whose code is on GitHub.
Codex told me the repo is missing training modules and data.
They claimed to open source it but left the core parts out.
An issue was opened by someone in June but got ignored.
How did a reviewer judge a CV paper? They just scan it and run no code?
If you check systems papers from OSDI/SOSP, reproducibility is very important. They usually open source the full codebase with a detailed guide for reproduction.
I want to share a preview of something I've been working on called Exact. It's a new Agent Native app framework written in Rust. It's extremely fast and tries to let you build exactly the experience you want with no compromises.
The ambient agent feature is released in Whip v2.1.0.
There are 3 modes of voice call: local, Realtime API and GPT-live.
The one using the Openai Realtime API has both low latency and VAD(it does not charge you when you are silent).
Please give it a try and advices are welcomed 😁