Chris Henson (@_chenson__) header
Chris Henson (@_chenson__) profile picture

CS PhD student @DrexelCCI working on formalization in Lean. Maintainer of https://teapawt.net/t.co/waYLdE0QGI. Student Researcher @GoogleDeepMind

Philadelphia, PA
Joined December 2014
Photo by Chris Henson (@_chenson__)Photo by Chris Henson (@_chenson__)Photo by Chris Henson (@_chenson__)Photo by Chris Henson (@_chenson__)Photo by Chris Henson (@_chenson__)Photo by Chris Henson (@_chenson__)Photo by Chris Henson (@_chenson__)Photo by Chris Henson (@_chenson__)Photo by Chris Henson (@_chenson__)
I figured my math followers would especially appreciate this! Just got these, they are originals from 1740 and 1751. Include the original publication of Euler's solution to the Basel problem, among other very cool things. So crazy to have these pieces of history on my bookshelf.
3
5
3
51
These folks have done a wonderful job of quickly setting up a mirror of the GRH formalization with initially just required setup changes and infra. There's a community effort for both improving the code and donating compute, please check the Zulip thread if interested in helping!
1
1
5
424
Chris Henson retweeted
To clear up some of the confusion:
15
48
3
1,432
39,370
Chris Henson retweeted
OK everyone, it's time to talk about the real part of every non-trivial zero of the Riemann zeta function (1/2)
80
501
60
5,221
208,290
I am in complete agreement about the dual nature of proof comprising both correctness and understanding. At the same time, AI-generated math is unlikely going anywhere, and we have to think about how to preserve these values in light of that.
I want to be clear about what my objection is here before all the idiots come after me saying, "Well, he posted a proof in Lean." Proofs are supposed to make math claims intelligible to others in a way that adds to math knowledge. Publishing a giant Lean blob doesnt meet that bar
5
14
1,531
There has been speculation that for GRH and unforced NS that the Lean formalization was not produced afterwards, but in a first-class role during development. Especially if this is true, it seems more important now than ever to think about how we can automatically informalize.
1
53
I don't really have great intuition for how this can be possible. My favorite quote to remember here is that "one can't proceed from the informal to the formal by formal means" and I think this applies. Still, can we approximate this? I think it's worth trying!
48
You'll never guess who this question is directed at...
Replying to @AcerFur
But do you understand *any* of it? Do you even know what an L-function is, say?
118
It is unconscionable to shove the Lean formalization for GRH into a dozen tar files in a footnote. Insisting on these sloppy presentation stunts for such important work is a disservice contrary to every bit of careful work aimed at responsibly using AI in math and formalization.
Replying to @scottnarmstrong
Footnote 1 of the pdf apparently?
35
17
5
443
44,391
This is exactly my long-term concern. After the initial wave of enthusiasm for formal methods, what does long-term sustainability look like? What happens when a piece of "verified software" later has a critical bug? Managing these comms in industry settings will be critical.
自分はもう少し悲観的で、ここ数年以内に来るだろう幻滅期を上手く乗り越えられなかった場合、DDD とか Kubernetes みたいな「賢しらな技術オタクが無理矢理導入して現場が困るあるある話」枠にされてしまいそうな予感を感じています。
1
1
7
840
roon talking about formal verification, I think it's undeniable we've hit the mainstream (with all the good and bad that entails)
imo this is confused thinking; you can automate safety research while being able to verify the results and therefore not having handed off human oversight. in fact, our powers of formal verification will only grow the better we understand these systems
1
20
901
Very sad to see comments like this. Work on the human-curated formalization is certainly not "done". Besides the fact that the proof they're formalizing is a modern version, there is certainly value to investing time in a higher quality of formalization than LLMs can provide.
Imagine if your grant becomes a benchmark for LLMs before you complete it lol
4
18
807
ICYMI, this part of the Fable 5.2 release has its own report: claude.dev/blog/safely-scali… This a a second full formalization of FLT, taking fewer tokens, a much more compact formalization, and 1000 agents figuring out on their own what to do without a tool like Prove2me.
Replying to @claudeai
A team of 1,000 Fable 5.2 agents fully formalized Fermat’s Last Theorem in 30 hours, translating the proof into a language a computer can check step by step. With the same setup, Opus 5.5 completed only 26% in 18 hours, and barely made any more progress after 30 hours.
2
9
525
Quite refreshing to see the long history of PL explicitly acknowledged by a startup doing AI for formal verification.
Formal methods didn't start with us. @ClarkBarrett7 on the people and tools &Truth builds on, from Turing, Hoare, and Dijkstra to Lean, Separation Logic, and SMT solvers: andtruth.com/blog/shoulders-…
2
22
1,463
Chris Henson retweeted
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. 🧵
27
34
9
364
49,847