Witold Wydmański
@wwydmanski.bsky.social
PhD candidate @ GMUM UJ. Working on tabular DL and domain-specific IR.
CTO of Gaius-Lex, a conversational search engine for lawyers
CTO of Gaius-Lex, a conversational search engine for lawyers
Take a look!
arxiv.org/abs/2503.13304
arxiv.org/abs/2503.13304
GFSNetwork: Differentiable Feature Selection via Gumbel-Sigmoid Relaxation
Feature selection in deep learning remains a critical challenge, particularly for high-dimensional tabular data where interpretability and computational efficiency are paramount. We present GFSNetwork...
arxiv.org
March 19, 2025 at 11:17 AM
Take a look!
arxiv.org/abs/2503.13304
arxiv.org/abs/2503.13304
I'd think that o1 would use ~as much energy per token as GPT4o, but produce much much more of them
January 13, 2025 at 2:59 PM
I'd think that o1 would use ~as much energy per token as GPT4o, but produce much much more of them
I'm still benchmarking it, but I think it should do much better than the base model
December 22, 2024 at 7:12 AM
I'm still benchmarking it, but I think it should do much better than the base model
AFAIK ModernBERT wasn't trained to have meaningful distances between vectors, take a look at tomaarsen/ModernBERT-base-gooaq
December 21, 2024 at 8:39 PM
AFAIK ModernBERT wasn't trained to have meaningful distances between vectors, take a look at tomaarsen/ModernBERT-base-gooaq
The trees are after you, watch out
December 14, 2024 at 6:51 PM
The trees are after you, watch out
Let me join you in your obsession, this is awesome
December 14, 2024 at 7:53 AM
Let me join you in your obsession, this is awesome
it do be like that
December 11, 2024 at 11:34 AM
it do be like that
Fair, it depends on the use case. Do you know if they're using it for autonomous tasks or for human assitance? I was thinking mostly about the former
December 7, 2024 at 5:10 PM
Fair, it depends on the use case. Do you know if they're using it for autonomous tasks or for human assitance? I was thinking mostly about the former
For the newest Llama, though, I fully agree
December 7, 2024 at 4:52 PM
For the newest Llama, though, I fully agree
o1 isn't available through the API yet, and even if it was it's prohibitively expensive for most of the use cases
December 7, 2024 at 4:52 PM
o1 isn't available through the API yet, and even if it was it's prohibitively expensive for most of the use cases
I do believe that is true
November 27, 2024 at 11:12 AM
I do believe that is true
Your DMs are closed, so 🙋
November 25, 2024 at 7:37 AM
Your DMs are closed, so 🙋
It's not related to your question at all, but every time I see "find" and "publication" in the same sentence I feel the need to talk about elicit.com
November 25, 2024 at 7:26 AM
It's not related to your question at all, but every time I see "find" and "publication" in the same sentence I feel the need to talk about elicit.com
Hi, could you add me to the feed? Here's my profile
scholar.google.com/citations?us...
scholar.google.com/citations?us...
November 22, 2024 at 8:04 PM
Hi, could you add me to the feed? Here's my profile
scholar.google.com/citations?us...
scholar.google.com/citations?us...