Yaron (Ron) Minsky (@yminsky) header
Yaron (Ron) Minsky (@yminsky) profile picture

Occasional OCaml programmer. Host of @signalsthreads. @yminsky@discuss.systems @yminsky.bsky.social https://xitter.twitterapi.xyz/t.co/kiUGRvWOO2

Joined June 2009
Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)Photo by Yaron (Ron) Minsky (@yminsky)
Yaron (Ron) Minsky retweeted
Self recommending! Come see me yap about straight lines on graphs.
I got to host a great panel on AI timelines: @DKokotajlo, @EgeErdil2, @Jsevillamol, and @RyanGreenblatt. It was an interesting event, a kind of follow-up to the old FOOM debate we hosted 15 years ago!
1
39
1,638
I got to host a great panel on AI timelines: @DKokotajlo, @EgeErdil2, @Jsevillamol, and @RyanGreenblatt. It was an interesting event, a kind of follow-up to the old FOOM debate we hosted 15 years ago!
2
3
2
61
4,404
Here's the link to the actual video: youtu.be/IlgvRtTwaUs?si=gYX-… It was interesting to me that there was a pretty big range of expected timelines for AGI amongst the panelists, from 2 years to 20 years. But at the same time, they all thought it was coming!
3
20
2,251
Yaron (Ron) Minsky retweeted
💭 unpopular opinion: type theories of Lean / Rocq are not complex and heavyweight enough. we need low-level, type-checked control over runtime *and* programs and proofs in the same lang maybe dependently typed lang with OxCaml's modes. even better: graded modal dependent types!
Lean, Rocq, etc. seem very complex and heavyweight for many applications but they serve as least common denominators, and now that AI can write in type theories, why not? Still, it feels like a smell and I sense something different is around the corner, not sure what though
23
15
1
175
8,099
Yaron (Ron) Minsky retweeted
Optimize kernels with a Python-based DSL and tile-centric primitives to automatically target Tensor Cores and other advanced GPU hardware features. CUDA Tile can take direct PyTorch implementations that compute multiple terms separately, fuse them into one tiled kernel and keep that matrix in registers. This blog shows how it enables you to write clear, readable, and highly performant kernels, including examples, docs, and sample code. Read the full post: bit.ly/3VigiKf
3
15
135
12,406
This is a pretty central question! One direction we're taking is to integrate specifications into OxCaml, so the spec language is easy and familiar for us. @JulesJacobs5's slides point to some of the directions we're trying. lab.julesjacobs.com/vox/talk…
Deciding how to help non-expert humans engage with formal verification specifications seems like one of the more important non-verifiable UX problems facing the rise of formal verification. 🧵
9
26
1
253
17,978
Some fun history of OCaml in places you might not expect.
Replying to @HeapSmasher
Our priority when building Docker for Mac (now Docker Desktop) was to make it feel transparent / like a native process even though Linux ran in a VM. Maybe this worked too well! I wrote about the history of our use of microVMs here: dave.recoil.org/docker-has-a…
1
7
104
7,141
Yaron (Ron) Minsky retweeted
ok i'm really geeking out now but it's very cool having a zero allocation type-safe OxCaml service using io_uring so almost no syscalls. I'm just having a go at replacing the other services with OCaml variants now (like the dhcp client)
2
12
63
3,187
Yaron (Ron) Minsky retweeted
I got so pissed off with our Netgear wifi routers running a vulnerable Linux from 2018 (!) that I've now got a custom Openwrt with Linux 7.3 running an OxCaml webserver/capnp. Better wifi performance with a new kernel too, and easy CLI management...
2
4
70
3,071
Yaron (Ron) Minsky retweeted
OxCaml 5.4.0+ox9 was released yesterday, and has a number of improvements to its performance-oriented extensions to OCaml. I've spent a few hours upgrading my infra and diving into details of the changes, and seeing what's in the pipeline... anil.recoil.org/notes/oxcaml…
4
22
1
124
6,075
Lots of cool things brewing...
Some news: I’m stoked to be joining the new Formal Methods team at Jane Street! Now is an amazing time for formal methods, with AI both empowering them and making them more necessary than ever. This is going to be fun!
5
4
1
338
21,116
Yaron (Ron) Minsky retweeted
A Canticle for Leibowitz would make such a great big budget Apple TV adaptation. You could probably get a solid seven seasons out of the general story and world
1
1
770
Yaron (Ron) Minsky retweeted
Crazy formal methods: water taxi to Spetses for a keynote at LPAR, then to DARPA ISAT in Arizona, and finishing with a quick stop at Jane Street, a proprietary theorem-proving shop, market-maker (for formal models and lemmas), and liquidity provider (for tokens and t-shirts).
2
1
43
2,982
Great thread
So the log_softmax bug arxiv.org/abs/2610.31642 that's been making the rounds. TorchLean also had the same problem (fixed in github.com/lean-dojo/TorchLe…). How come? This example is pretty illustrative of the proper mindset you need when working on formal verification. 🧵
1
85
12,102
Love seeing ideas hop languages!
Inspired by oxcaml's zero_alloc, I've been working on a tool to help clamp down on heap allocations in Golang. github.com/theconsensuslabs/…
1
44
3,736
Brilliant
With the way things are going Captchas are going to ask users to design a UI without adding unnecessary UI text.
17
2,533
Highlights the importance of picking the right hill to climb on.
For what it's worth, the metric I'm measuring is: take a few real Codex sessions, look at all the file edits and Cargo commands, and then replay those. How much time is saved end-to-end?
1
1
64
7,591
This seems like an excellent metric
For what it's worth, the metric I'm measuring is: take a few real Codex sessions, look at all the file edits and Cargo commands, and then replay those. How much time is saved end-to-end?
47
9,540