halilakin
@halilakin
Followers
1K
Following
2K
Statuses
267
Always tries to lower entropy, sometimes generates novel data. Scaling @evoscaleai
Joined May 2009
The hottest new programming language is now biology 😉 #esm3
The hottest new programming language is English
0
3
52
@DimitrisPapail What do you mean "common fingerprint"? Could you give any examples to illustrate the concept?
1
0
1
Meanwhile, at Big Tech: seven people are in a meeting about the work one person is actually doing, followed by six complaining about how difficult coordination is and how they don’t have time to get any work done.
Last tweet on this but the way @deepseek_ai does launches is beautiful: no hype, arrogance or vague-posting: just sharing something great with the world. US tech companies look cringe in comparison.
1
0
8
RT @ScienceMagazine: An #AI model created to design proteins simulates 500 million years of protein evolution in developing a previously un…
0
247
0
RT @alexrives: We're thrilled to present ESM3 in @ScienceMagazine. ESM3 is a generative language model that reasons over the three fundamen…
0
252
0
RT @salcandido: We built new LLMs. 1/ Scaling curve moved up and to the left(!!) since ESM2 with the new trainouts. 2/ Inference efficienc…
0
6
0
RT @rohilbadkundri: Check out ESM Cambrian, our latest models focused on representation learning! We found scaling compute and data to the…
0
5
0
RT @TomSercu: Announcing ESM Cambrian. ESM C defines a new state of the art for protein sequence modeling. ESM C is a drop in replacement…
0
11
0
RT @velvetatom: Thrilled for first family of ESM models, ESM-C 6B to launch for commercial biopharma use with open weights for ESM-C 300 &…
0
11
0
RT @deaton_jon: This is an incredibly cool plot - it shows how protein language models form internal representations of physical structure…
0
2
0
RT @sofroniewn: Super excited about ESMC and the quality of the protein representations! Can't wait to see what people build on top of it
0
1
0
No walls to hit here in biology! ESM C is our compact representation learning model family, delivering dramatic improvements over ESM2.
Information about protein structure in ESM C representations improves predictably with increasing training compute, demonstrating linear scaling across multiple orders of magnitude. (We overtrained the 300M and 600M models past the predicted point of compute optimality).
0
1
20
RT @alexrives: Introducing ESM Cambrian. Unsupervised learning can invert biology at scale to reveal the hidden structure of the natural w…
0
175
0