π₯Introducing Gemini 2.5, our most intelligent model with impressive capabilities in advanced reasoning and coding.
Now integrating thinking capabilities, 2.5 Pro Experimental is our most performant Gemini model yet. Itβs #1 on the LM Arena leaderboard. π₯
25.03.2025 17:25
π 215
π 65
π¬ 34
π 11
Protein waste turned into antibiotics as a defence strategy of human cells
Proteasome cleavage changes during infection to generate antibiotics.
Proteasomes make natural antibiotics while breaking down proteins. So cool to see research in new areas revealing another layer of biological complexity. Makes you wonder what else is out there! www.nature.com/articles/d41...
10.03.2025 10:28
π 3
π 0
π¬ 0
π 0
Slashed zero (0ΜΈ), a representation of the number 0 (zero) to distinguish it from the letter O
Slashed letter 'O' (OΜΈ), a representation of the letter O to distinguish it from the number 0 (zero)
19.02.2025 06:57
π 14191
π 2709
π¬ 193
π 236
wait itβs not hopeless?
13.02.2025 23:24
π 1
π 0
π¬ 0
π 0
Jeff Dean & Noam Shazeer β 25 years at Google: from PageRank to AGI
YouTube video by Dwarkesh Patel
Delighted to have joined my good friend & colleague Noam Shazeer on a podcast with @dwarkesh.bsky.social for a 2+ hour discussion on early Google, ML hardware, training 1T+ token LLMs in '07, model sparsity, continual learning, and more.
Thanks, Noam and Dwarkesh! π
youtu.be/v0gjI__RyCY?...
12.02.2025 21:59
π 97
π 13
π¬ 1
π 3
Reading group session on Monday: Valentin De Bortoli joins us for his paper "Accelerated Diffusion Models via Speculative Sampling" arxiv.org/abs/2501.05370
On zoom at 9am PT / 12pm ET / 6pm CET: portal.valencelabs.com/logg
01.02.2025 19:38
π 14
π 2
π¬ 0
π 0
super cool paper!
17.01.2025 13:40
π 2
π 0
π¬ 1
π 0
A month ago we @vevotherapeutics.bsky.social announced that we have generated the largest single-cell perturbation atlas in history, Tahoe-100M. Today, we announce that we will fully open-source Tahoe-100M in Feb, as part of a collaboration with NVidia health to train cell state models.
13.01.2025 16:23
π 116
π 33
π¬ 4
π 5
OpenADMET Jobs
Seeking talented scientists for the OpenADMET Consortium
Interested in building the future of open source ligand- and structure-based ML models for ADMET prediction?
The ARPA-H funded OMSF (@omsf.bsky.social) OpenADMET project is hiring multiple positions to build ML models of ADMET properties and drive informative data collection!
openadmet.org/jobs/
08.01.2025 15:41
π 37
π 13
π¬ 0
π 1
Features of Adulthood xkcd.com/3034
06.01.2025 23:14
π 6694
π 1023
π¬ 116
π 93
52 things I learned in 2024
This year I took a sabbatical, recorded a podcast with Freakonomics, designed and shipped a tiny modular synthesiser, and learned many learnings. Tom Whitwell designs open source music electronics asβ¦
This yearly list is fantastic as usual: medium.com/@tomwhitwell...
For example:
> To highlight tax evasion, South Korea introduced ugly neon green number plates for company cars worth more than $58,000. Luxury car sales fell 27%. [Song Jung-a]
19.12.2024 15:12
π 4
π 0
π¬ 1
π 0
so cool, congrats π
12.12.2024 21:46
π 1
π 0
π¬ 0
π 0
#TeamTomo is at it again. Nice look at the conformational cycle of protein folding machinery inside native human cells by the teams of Frydman & Beck π§ͺπ¬π§Άπ§¬
11.12.2024 18:57
π 22
π 4
π¬ 1
π 0
tbf it doesnβt seem particularly strange - i glanced at the screenshot after reading this and it looks like the βtop10 most popularβ section of an average bookshop.
11.12.2024 10:29
π 1
π 0
π¬ 0
π 0
Lots coming for protein design!
08.12.2024 10:33
π 4
π 1
π¬ 0
π 0
If you will be at #NeurIPS2024 @neuripsconf.bsky.social and would like to come see our models in action, come say hi π and check out our demo at the GDM booth!
Wednesday, Dec. 11th @ 9:30-10:00.
Lots of other great things to see as well! Check it out: π
deepmind.google/discover/blo...
06.12.2024 12:42
π 41
π 6
π¬ 1
π 0
Excited to reveal Genie 2, our most capable foundation world model that, given a single prompt image, can generate an endless variety of action-controllable, playable 3D worlds. Fantastic cross-team effort by the Open-Endedness Team and many other teams at Google DeepMind! π§
04.12.2024 16:13
π 97
π 18
π¬ 3
π 3
fun times!
05.12.2024 21:56
π 1
π 0
π¬ 0
π 0
Fair enough. I guess we'll all scramble until we find an acceptable one. It takes a while to find out what kind of interview leads to the fairest process and the best hires. Probably back to having more face time?
05.12.2024 09:00
π 1
π 0
π¬ 1
π 0
Some techs and modalities are massively over-represented so it becomes hard to determine if it's a problem that would stop generalization.
Reprocessing most single cell data in a single pipeline would allow to diagnose this better (and would be a phenomenal service to the sc community!)
04.12.2024 10:14
π 1
π 0
π¬ 1
π 0
Hackerrank was broken - but now it's actually harmful β segfaulte
constructing good interviews is a never ending quest - here is a quick read on how access to LLMs changes incentives segfaulte.mataroa.blog/blog/hackerr...
04.12.2024 10:08
π 3
π 1
π¬ 2
π 1
as a kid, i fixed my parent's computer. as an adult I fix my kids's computer. pretty sure we are the only gen that knows how computers work
03.12.2024 12:37
π 11968
π 925
π¬ 378
π 105
hope one day to see an animation of the clockwise to counterclockwise switching mechanism :)
02.12.2024 10:22
π 2
π 0
π¬ 0
π 0
Taming Transformers for High-Resolution Image Synthesis
Designed to learn long-range interactions on sequential data, transformers continue to show state-of-the-art results on a wide variety of tasks. In contrast to CNNs, they contain no inductive bias tha...
IMO VQGAN is why GANs deserve the NeurIPS test of time award. Suddenly our image representations were an order of magnitude more compact. Absolute game changer for generative modelling at scale, and the basis for latent diffusion models.
28.11.2024 00:09
π 104
π 15
π¬ 2
π 2
Promotional logo for MMseqs2 16 with the MMseqs2 Rocket mascot as a smart phone like App logo
MMseqs2 Release 16 Highlights: GPU-accelerated searchπ, ORF or new 6-frame translated search modes, contig taxonomy always keeps the longest ORF, bug fixes (reduced memory and higher sensitivity) and relicensed as MIT
π biorxiv.org/content/10.1...
πΎ mmseqs.com and πBioconda π₯οΈπ§¬π§Ά
27.11.2024 09:08
π 115
π 44
π¬ 0
π 1