@DiracDeltaFunk

Sheaf herder. I believe in you 🔥

 Bloomington
benspitz.com
Joined October 2013
  • Tweets5.6K
  • Following693
  • Followers2.7K
193Media
I did a vibecoding. Announcing *bibers*, a drop-in replacement for biber (the BibLaTeX backend processor) which is much faster. You can install it from homebrew, and it's easy to set up your tools to use bibers instead of biber. github.com/diracdeltafunk/bi…

GitHub - diracdeltafunk/bibe-rs: A fast, biber-compatible bibliography processor for biblatex, in...

A fast, biber-compatible bibliography processor for biblatex, in Rust - diracdeltafunk/bibe-rs

github.com
2
10
739
This was basically a one-shot by Opus 5.5. FAQ: Q. You had opus write hundreds of thousands of lines of code to... improve your LaTeX compile times by a couple seconds? A. Yeah. I hope it's useful to you as well!
1
123
You know how "prime ideal" is annoyingly a conjunction of two conditions? (1) ab ∈ P ⇒ a ∈ P or b ∈ P (2) 1 ∉ P These can be combined!
7
1
2
121
21,442
An ideal P of a commutative ring R is prime iff: For all finite S ⊆ R, if ∏S ∈ P then S∩P ≠ ∅. --- Take S = ∅ to get (2)!!!
4
2
79
3,263
Ben Spitz (63/100 improv meals) retweeted
Quantіan
@quantian1Oct 7
Replying to @RadishHarmers
There's no point in coping. The AI will be better than you at coping within 6 months.
13
200
11
2,214
38,354
No longer open, it seems
Ben Spitz (63/100 improv meals)@DiracDeltaFunk20 Aug 2025
An open problem in order theory: Let L be a finite lattice (i.e. a nonempty finite poset such that any two elements have both an inf and a sup). Must there exist a finite group G with subgroups H, K such that L is isomorphic to the poset {X ≤ G : H ≤ X ≤ K} ordered by ⊆?
2
1
55
3,214
"Average person makes 2 restaurant reservations per day" factoid actually just statistical error. Average person makes 0 restaurant reservations per year. Restaurants Georg, who lives in SF and makes over 10,000 reservations per day, is an outlier adn should not have been counted
5
492
Muse is a very good product, ngl. The voice input is almost flawless -- importantly, it perfectly edits our any "um"s and stammers before sending the text to the model. Makes a huge difference for usability.
1
1
368
Worth saying that the model sometimes makes ridiculous mistakes and is pretty bad at interacting with the internet. Nothing disastrous so far, but it's just barely on the edge of feeling safe enough to use. It's CRAZY that people were doing this (clawdbot) with models 1yr ago!
124
Ok now that things have quieted down a little bit, is there something good I can read to figure out what's actually up with the fly connectome? The "I made a fly DJ a party" posts seem obviously somewhat fake/not really descriptive of what's happening.
4
12
965
more replies
Idk, maybe I'm ignorant of something. But it's like if someone uploaded just the floor plan of an alien power plant, and then everyone was like "woah I just simulated what would happen if they used molasses instead of water as coolant!" ??? no, you are missing key information.
1
3
177
Apologies for the very bad analogy, I'm so confused about the situation that I can't seem to find a good one lol. But hopefully this communicates something.
2
91
It is surprising to me how bad most benchmarks seem to be. Humanity's last exam is like 50% wrong??? How is it possible that no one is catching these many many errors before the benchmark is published?
Epoch AI
@EpochAIResearchSep 17
Introducing Benchmark Reviews: our new initiative to audit AI benchmarks. We are launching with 15 benchmarks: 4 Verified, 9 Flawed, and 2 with not enough information for a review.
18
2
3
247
17,186
Seriously, look at the report on HLE. It's crazy.
18
1,229
Pacing the frontier
Sam Altman
@samaSep 16
the main thing i was excited about launching this week will be next week instead, but imo worth the wait!
1
412
Ben Spitz (63/100 improv meals) retweeted
Leonardo de Moura@Leonard41111588Sep 10
Lean has a new checker: con-leche, a CONsistent LEan CHEcker. This is an external checker for Lean that is proven (in Lean) to be consistent, meaning it does not accept a proof of False. Joachim Breitner (@nomeata) is the mastermind behind the project. github.com/leanprover/con-le…

GitHub - leanprover/con-leche: An external lean checker with a proof of consistency

An external lean checker with a proof of consistency - leanprover/con-leche

github.com
19
77
20
551
79,597
A surprisingly nice use case for small models, for me, is as a magic oracle that gives me unicode symbols to copy & paste. On a related note, I switched to macOS a few years ago and really miss WinCompose :(((
2
215
Do you think it is possible to "solve alignment"?
17%yes
42%no
14%idk
27%what does this mean?
100 votes • Final results
4
5
782
I would be very excited to see a resolution of P vs NP, even if it's slop, ngl
2
13
902
Load more