Chris Henson

4.7K posts

Chris Henson banner
Chris Henson

Chris Henson

@_chenson__

CS PhD student @DrexelCCI working on formalization in Lean. Maintainer of https://t.co/waYLdE0QGI. Formerly Quant Finance at Bank of America.

Philadelphia, PA Katılım Aralık 2014
2.7K Takip Edilen1K Takipçiler
Sabitlenmiş Tweet
Chris Henson
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.
Chris Henson tweet mediaChris Henson tweet mediaChris Henson tweet mediaChris Henson tweet media
English
3
5
52
0
danny
danny@danny_nkjg·
@_chenson__ @MarioKrenn6240 @mathematics_inc Thanks Chris!! So this particular tweet/quote is more of a reaction to low quality slop generations, and less about fears about math research becoming more AI driven?
English
1
0
2
144
Mario Krenn
Mario Krenn@MarioKrenn6240·
After the apparently amazing announcement by @mathematics_inc on the formalization of a major recent Fields-medal winning theorem, i had no idea how pissed the math-formalization community is. Very worrying discussions by some of the leaders/founders of Lean's mathlib. cc @ChrSzegedy
Mario Krenn tweet media
English
22
43
547
190.8K
Kevin Hartnett
Kevin Hartnett@KSHartnett·
In my book "The Proof in the Code" I have several chapters devoted to the birth and growth of Mathlib. The maintainers (who determine whether a PR is admitted to the library or not) are extremely exacting: they don't just want definitions and theorems formalized; they want them formalized in an elegant, maximally general way that can serve as the right building blocks for all future results that would draw on them. These exacting standards are why Mathlib has been so successful, and they're why AI autoformalization that isn't to the same standard is so threatening. Pre-orders: quantabooks.org/books/the-proo…
ludwig@ludwigABAP

wow wow wow holy BASED I love Patrick Massot and I have been saying that for a year

English
4
18
192
17.5K
Chris Henson
Chris Henson@_chenson__·
@danny_nkjg @MarioKrenn6240 @mathematics_inc In short: giant AI developments with unclear value because quality issues make them impossible to reuse. These developments have multiple times, to my understanding without coordination, been aimed at ongoing projects producing much higher quality work suitable for Mathlib.
English
1
0
2
131
ludwig
ludwig@ludwigABAP·
I will pre-ordering because I have tried to piece together some of the lore behind the birth and growth of Mathlib from my own readings but do not know the exact story, so that is brilliant. cc @crypt0lake @basedmachina @cachecrab I believe I discussed this quickly with each of you in a space / IRL
English
2
0
21
1.8K
Daniel Litt
Daniel Litt@littmath·
@ludwigABAP To be honest I think there’s some chance this happens to research math as a whole, if we don’t adapt. It’s possible we end up with an equilibrium where the tools are worse than human mathematicians but good enough to “justify” massive cuts and loss of human capital.
English
11
16
286
11.8K
ludwig
ludwig@ludwigABAP·
Everyone who is currently bombared by "solve math solve everything" "autoformalization -> AGI" psyops should read this and try to understand where this sentiment comes from
ludwig tweet media
English
15
8
283
53.5K
Chris Henson
Chris Henson@_chenson__·
@joseph_h_garvin It is a bit ambiguous, but I'd say the 1930s as shared origins with mathematics that colors PL research. Things have obviously changed a lot since then and software engineering culture doesn't make sense to discuss until later, but I think of this as a branching point.
English
0
0
19
833
Chris Henson
Chris Henson@_chenson__·
@Keleesssss In Lean they also (since before modern AI) have a doc header for every module that lists primary authors for attribution/maintenance purposes, so I think there's less attachment to commit coauthors. Little details like this make a big difference in the social reception I'd guess
English
0
0
2
32
Alperen Keleş
Alperen Keleş@Keleesssss·
@_chenson__ I feel like there are better alternatives for disclosure purposes though. I’ve been reading discussions in Agda and Idris communities, the commit authorship is diverting the discussion unnecessarily into who owns/takes responsibility for the code.
Alperen Keleş@Keleesssss

@redixhumayun I personally hate it. Commit authorship (to me) is about responsibility and knowledge, it’s who you go to when something breaks or when you’re trying to understand something. I would be fine seeing only Claude for autonomous commits, this version just muddies the waters.

English
2
0
2
167
Alperen Keleş
Alperen Keleş@Keleesssss·
On a separate note, is there any sensible explanation of why Claude is added as an author on a commit? Codex doesn’t do it, so as I understand Anthropic encoded this behavior specifically into their model/harness. Why? What’s the benefit to anyone? What’s the point?
Yuchen Jin@Yuchenj_UW

OpenAI acquired Astral, the team behind uv, ruff, and ty. Fun fact: Claude is the #6 contributor to uv. Curious if Anthropic will ban them from using Claude since the team is joining OpenAI. Congrats to the Astral team who built incredible Python tools!

English
4
0
18
5.5K
Erik Meijer
Erik Meijer@headinthebox·
Question for the proof assistants crowd. In Lean/Rocq/Agda the programming language and the theorem proving tool are conflated under one name. "Lean" is both the dependent type theory and the elaborator/tactic framework/kernel. There's no concept of "Lean the language could have a different runtime." Why is that? To me it sounds like conflating an interface with its implementation. Or am I just being a dimwit?
English
14
0
53
8.3K
Chris Henson
Chris Henson@_chenson__·
It's interesting how we now have domain specific AI tells. When someone opens with "no axioms, zero sorry" about a Lean proof, it is almost guaranteed to be AI-generated.
English
0
0
7
298
Chris Henson
Chris Henson@_chenson__·
@AcerFur @mathematics_inc Can you clarify exactly what autoformalised means here? In previous announcements this has elided important details of exactly what inputs are involved.
English
1
0
1
253
Chris Henson
Chris Henson@_chenson__·
@lydiahallie @Sauers_ @AnthropicAI I think the flexibility as listed is great! Just wanted to mention formalization specifically because it has some distinctions from most software.
English
0
0
1
74
Sauers
Sauers@Sauers_·
Terence Tao "ran out of tokens" on Claude Code, causing the formalization task he was working on to fail. @AnthropicAI should probably give him free Pro lol
English
36
33
2K
134.5K
Chris Henson
Chris Henson@_chenson__·
@lydiahallie @Sauers_ @AnthropicAI I hope especially with formalization that some consideration will be given to its niche nature and relevance to AI safety. Not even Mathlib meets the criteria listed!
English
1
0
6
768
Lydia Hallie ✨
Lydia Hallie ✨@lydiahallie·
@Sauers_ @AnthropicAI ah that's fine, we also review people who don't meet the criteria but are still doing important/invisible work :)
English
3
0
24
1.3K
Chris Henson
Chris Henson@_chenson__·
@bcubeddd Just to clarify, the intro says that "there was no extensive use of AI in the formalization within this paper", right? I have some serious issues with misleading claims of "autoformalization" in general, but this seems like a separate situation.
English
1
0
1
50
big brane boi
big brane boi@bcubeddd·
I anticipate that less known authors will have their results autoformalised and “falsified”, while more famous ones will escape scrutiny as always
English
2
0
6
499
big brane boi
big brane boi@bcubeddd·
original research is hard and yes, mistakes may happen. reheating 20yr old nachos topped with the flavour of the month autoformalisation is much easier
big brane boi tweet media
English
4
2
31
3.4K
Chris Henson
Chris Henson@_chenson__·
@JustDeezGuy In 2020 I got the worst of both worlds: I'd just moved halfway across the country to Atlanta then ended up working remotely for over a year at a 5 minute walk from our office in midtown. It was pretty tough both as a brand new junior and as someone who knew nobody in the city.
English
0
0
2
113