Win Wang
857 posts
Win Wang
@Winium
Founder @ ParaQuery (YC X25) | Data processing for the GPU era. Dilettante of FP, PL, philosophy, physics, quantum computing. Possibly writing code.
Beigetreten Haziran 2011
786 Folgt396 Follower
@Winium And in this talk, I'll be like "So here's what I was going to tell there", and just proceed with my original talk.
Clever.
English
Win Wang retweetet
We doubled Claude usage on weekends, and outside 5–11am PT on weekdays for the next 2 weeks.
Claude@claudeai
A small thank you to everyone using Claude: We’re doubling usage outside our peak hours for the next two weeks.
English
Win Wang retweetet
@lydiahallie Amazing. I did this all the time as a custom skill-like solution, glad to see it formally in CC!
English
You can now ask Claude a quick question about your current session without interrupting the main task!
It's read-only (no tool access) and doesn't add to the conversation history. The answer just disappears when you dismiss it
Thariq@trq212
We just added /btw to Claude Code! Use it to have side chain conversations while Claude is working.
English
I predict in the near future,
the majority of GPU cycles in the world will execute code written in Python DSLs that lower to machine code without going through C++.
Triton, JAX/Pallas, torch.compile, cuTile, CUTE DSL, etc
The era of C++ and C as the foundational layer is over.
lyv ⌘@wholyv
What if C++ was used for machine learning instead of Python. given C++ evolved to allow that. I wonder how faster, our AI models would be these days.
English
Win Wang retweetet
We’re excited to release TorchLean which is the first fully verified neural network framework in Lean. The Lean community has largely focused on pure mathematics. TorchLean expands this frontier toward verified neural network software and scientific computing. With the recent release of CSlib, we see this as another step toward a fully verified ML stack.
We support features:
1. Executable IEEE-754 floating-point semantics (and extensible alternative FP models) verified tensor abstractions with precise shape/indexing semantics
2. Formally verified autograd system for differentiation of NN programs Proof-checked certification / verification algorithms like CROWN (robustness, bounds, etc.)
3. PyTorch-inspired modeling API with eager-style development + export/lowering to a shared IR for execution and verification
Project page: leandojo.org/torchlean.html
Paper: [2602.22631] TorchLean: Formalizing Neural Networks in Lean
Work done @Robertljg, Jennifer Cruden, Xiangru Zhong, @huan_zhang12 and @AnimaAnandkumar.
#MachineLearning #ScientificComputing #Lean
English
@fbrasisil I've actually been working with different CLAUDE_CONFIG_DIR with multiple agents in the same repo dir. I use a commit mutex with a prompt "macro" for single-time auto-approved commits. Also working on a fully dockerized solution with tmux'd persistence.
English
@Winium yeah, the new built-in worktree feature isn't great. Sub agents don't respect it and file searches end up returning content from the main repo or other worktrees. I'm managing worktrees manually in separate folders. It works ok that way
English
to be fair: I complained about Claude Code recently but the beast is back! working on multiple parallel sessions is truly a next-level experience 😍
English
@headinthebox @fbrasisil Yes. I find that Claude writes significantly better Scala, not to mention the static guardrails Scala allows.
English
@fbrasisil Existential question, does the programming language even matter anymore when the AI agent does all the work?
English
A nice plot twist: Scala's bad DevEx is greatly mitigated by AI agents. I don't bother with broken IDE refactoring, don't need to fight sbt anymore, don't need to debug stuff because of bad docs, can easily pinpoint compiler bugs and find workarounds, etc. Just ship 🚀🚀🚀
English
Anthropic’s War on Its Own Power Users
Charging power users $200/month then banning them for actually using it is not ‘abuse prevention’ — it’s the RIAA playbook for AI, and we all remember how that movie ends.
garryslist.org/posts/anthropi…
English
@paulg Depends on the code, probably. Also, I think there was something like a massive upgrade around oct/nov last year. I've aleays been pro-AI but it was mostly unusable for my projects outside of being a great googler. Now, idk about 1k/hr, but definitely thousands/day.
English
An experienced programmer told me he's now using AI to generate a thousand lines of code an hour.
When I posted a similar stat 6 months ago, I got about a 50-50 mix of indignant disbelief and "Yeah, me too." I'm curious if the split will be different this time.
English