Sumuk (@sumukx) / X

Sumuk

1,314 posts

Sumuk

@sumukx

continual learning research @google / prev @PrimeIntellect @huggingface | opinions my own

San Francisco, CA

sumuk.org

Joined September 2023

Pinned
Sumuk
@sumukx
Apr 2, 2025
we're launching 🤗 yourbench today, an open source tool for custom benchmarking and synthetic data generation from ANY of your documents. it's a big step towards improving how model evaluations work early access link in replies! (1/8)
49K
Sumuk
@sumukx
Jul 7, 2024
Someone said that teachers hate testosterone and I don’t think they could be more right lol
35K
Sumuk
@sumukx
Jan 9, 2024
Replying to @growing_daniel
I feel like society, and the times were different. People trusted others easily. Also why hitchhiking was possible
21K
Sumuk
@sumukx
Jul 29, 2024
Replying to @chucky
Right - you’re still not responding to some of the worst ones lol
23K
Sumuk
@sumukx
Jul 30, 2025
Replying to @SIGKITTEN
the second it starts using emojis, i kill the session
14K
Sumuk
@sumukx
Aug 26, 2025
Nous Research makes some of the best tasting models ever. It's singular in how "alive" it feels. The only model that comes close is the original claude 3 opus. Best of all? It's open. I encourage you all to try it for yourselves. Nebius has a generous free tier.
12K
Sumuk
@sumukx
Jul 4, 2024
Replying to @krishnanrohit and @ValerioCapraro
You haven’t worked enough with ChatGPT to be able to tell. Delve into it a bit more.
10K
Sumuk
@sumukx
Jul 8, 2024
Replying to @mohbibi_
Dumb take. 120hz is literally a night and day difference it’s unreal
4.7K
Sumuk
@sumukx
Jul 15, 2025
Replying to @Yuchenj_UW
imo okay to release 8 - 405B, and maybe keep 2T to themselves, but abandoning is really really bad its a +1 to the OpenAI, Anthropic, Google list and its hard to differentiate
14K
Sumuk
@sumukx
Jun 22, 2024
Replying to @EsotericCofe
I feel like this is the Indian equivalent of GPT generated slop articles lul
14K
Sumuk
@sumukx
Aug 5, 2025
In just ONE line of code, you can use the new gpt-oss model to turn your messy raw data (pdf, word, xlsx) into a clean, strong eval set (to test any LLM!) with yourbench! (link in comments)!!
9K
Sumuk
@sumukx
Mar 12, 2025
Replying to @vikhyatk
man the fact that i even double looked at this 💀 do a walmart supercenter pro max imposed over africa
2.9K
Sumuk
@sumukx
Jul 8, 2024
They’re basically selling to the cope market, which wants to hear that AI is useless. They’ll be left in the dust when people with a competitive advantage due to stronger models become the norm. I don’t know why you think this is a bad thing. Those who know models, will know.
2.6K
Sumuk
@sumukx
Jun 12, 2025
Replying to @iamgingertrash and @swyx
the yc scaling law of “revenue scales exponentially with batch size”
5K