trott

7 ms·

1.
Tiny Model, Big Logic: Large-Model Reasoning Ability in VibeThinker-1.5B(arxiv.org)
4 points by trott 10mo ago|0 comments
2.
by trott 1y ago
> The AlphaFold3 analysis (the AI contribution) literally accounts for a few panels in a supplementary figure - it didn't even help guide their choice of small molecule inhibitors since those were already known. (Disclaimer: I'
3.
The flaw in cryonics that could ruin your shot at living forever(sciencefocus.com)
2 points by trott 1y ago|1 comments
4.
Is this the $200K ticket to cheating death?(bbc.com)
3 points by trott 1y ago|0 comments
5.
by trott 1y ago
Regarding point number 11 (AlphaFold3 vs Vina, Gnina, etc.), see my rebuttal here (I'm the author of Vina): https://olegtrott.substack.com/p/are-alphafolds-new-results-... Gnina is Vina with its results re-scored
6.
by trott 1y ago
> According to Stack Overflow developer survey [0] Rust is at 12.5%, ... So definitely not niche. The annual survey is very popular in the Rust community. Its results are often used for advocacy. Participation by Rust developers is very
7.
by trott 1y ago
They started with N >= 120x3 tasks, and gave each task to 4-9 humans. Then they kept only those 120x3 tasks that at least 2 humans had solved.
8.
ARC Prize side quest: SnakeBench(arcprize.org)
4 points by trott 2y ago|0 comments
9.
by trott 2y ago
> a mistake of the 1983 Ada standard ... which has been removed The article was written in 2011, and the trick still seems to work in a 2024 version of GNAT. > Both "unchecked" Ada and "unsafe" Rust But the `Conver
10.
by trott 2y ago
> but Ada's real strengths lie elsewhere. Its strong typing, Ada is not actually type-safe: https://www.enyo.de/fw/notes/ada-type-safety.html
11.
by trott 2y ago
> it will finally be fixed relatively soon 2015: "The work needed to close this has not yet landed. It's in the queue though, once we finish up rust-lang/rfcs#1214."
12.
by trott 2y ago
Yep, and even without dynamic memory management, Ada is not type-safe: https://www.enyo.de/fw/notes/ada-type-safety.html Rust also has soundness holes, by the way. This one is almost 10 years old: https:/&#x
13.
by trott 2y ago
Another way to look at this is that there are 12,290 bits of information in choosing 817 samples from 10,000,000.
14.
LIMO: Less Is More for Reasoning(arxiv.org)
389 points by trott 2y ago|128 comments
15.
De novo designed proteins neutralize lethal snake venom toxins(nature.com)
1 points by trott 2y ago|0 comments
16.
by trott 2y ago
How do Maxima and SymPy compare in terms of capability, features and speed (native, not WASM)?
17.
by trott 2y ago
This is the paper: https://arxiv.org/abs/2411.02272 They won the 1st paper award: https://arcprize.org/2024-results In their approach, the LLM generates inputs (images to be transformed) and solutions
18.
by trott 2y ago
> This has me curious about ARC-AGI In the o3 announcement video, the president of ARC Prize said they'd be partnering with OpenAI to develop the next benchmark. > mechanical turking a training set, fine tuning their model You do
19.
by trott 2y ago
> Terra programs use the same LLVM backend that Apple uses for its C compilers. Can it use anything else (as an option), e.g. Lua? That would be useful during development/debugging thanks to faster iteration and memory safety.
20.
Frontier Models are Capable of In-context Scheming(arxiv.org)
10 points by trott 2y ago|1 comments
21.
by trott 2y ago
Not sure. But I think this follows logically from the definition of intelligence he is using. Also, see II.2.2 in the paper.
22.
by trott 2y ago
> That's fine if you want to measure sample efficiency, but ARC-AGI is supposed to measure progress towards AGI. On the Measure of Intelligence defines intelligence as skill-acquisition efficiency, I believe, where efficiency is w
23.
by trott 2y ago
> realise that you're wrong, accept it, and move on What do you think about limiting the submission size? Kaggle does this sometimes. With a limit like 0.1-1MB (compressed), you are basically saying: "Give me sample-efficient l
24.
by trott 2y ago
(Spoiler alert) In https://arcprize.org/play?task=79369cc6 , the yellow 3x3 square shows you the pink pattern to look for, while allowing rotations (and ignoring the fact that the pattern may be next to other patterns)
25.
by trott 2y ago
Mike and François, Compute is limited during inference, and this naturally limits brute-force program search. But this doesn't prevent one from creating a huge ARC-like dataset ahead of time, like BARC did (but bigger), and training a
26.
ARC Prize 2024: Technical Report [pdf](arcprize.org)
2 points by trott 2y ago|0 comments
27.
Duct-taped banana artwork sells for $6.2M in NYC(bbc.com)
7 points by trott 2y ago|0 comments
28.
Test-Time Training on Nearest Neighbors for Large Language Models(arxiv.org)
1 points by trott 2y ago|0 comments
29.
by trott 2y ago
Congrats, François, and good luck! Q: The ARC Prize blog mentions that you plan to make ARC harder for machines and easier for humans. I'm curious if it will be adapted to resist scaling the training dataset (Like what BARC did -- see
30.
by trott 2y ago
> That's not general intelligence, that's gaming. In fairness, their approach is non-trivial. Simply asking GPT-4o to fantasize more examples wouldn't have worked very well. Instead, they have it fantasize inputs and progr

More ›