aaronw3r's profile picture. PhD AI for Drug Discovery @QMUL, @EPFL, @MSDintheUK. ex-{@TUDelft, @CureVacRNA}. Researching the intersection of AI and biology to improve human health. 🧬🤖

Aaron Wenteler

@aaronw3r

PhD AI for Drug Discovery @QMUL, @EPFL, @MSDintheUK. ex-{@TUDelft, @CureVacRNA}. Researching the intersection of AI and biology to improve human health. 🧬🤖

Pinned

1/14 Excited to introduce PertEval-scFM! It provides benchmark and evaluation tools for perturbation effect prediction models, including single-cell foundation models (scFMs). Paper and GitHub link at the end of the thread! 🧵👇

aaronw3r's tweet image. 1/14 Excited to introduce PertEval-scFM! It provides benchmark and evaluation tools for perturbation effect prediction models, including single-cell foundation models (scFMs). Paper and GitHub link at the end of the thread! 🧵👇

Fan of the blue light on the London night buses

aaronw3r's tweet image. Fan of the blue light on the London night buses

Can’t blame AI for this slop on my plate can I?

aaronw3r's tweet image. Can’t blame AI for this slop on my plate can I?

I’ll be hitting up SF and LA in the week leading up to NeurIPS. Eager to meet people in the area working on AI, biology and/or drug discovery. Let me know if you’ll be around at the end of November


Lessons from these "failures" led to groundbreaking results. Good research prioritizes uncovering replicable truths. In RL terms, exploration (Sutton's focus) and exploitation are interdependent. Any exploration that deepens our understanding meaningfully is not a negative result

Rich Sutton's career: 1. Do a lot of research on reinforcement learning. 2. Write an essay on why it failed. 3. Win the Turing award for the failed research.



Even as an applied researcher pretty far removed from this type of kernel engineering: this is very cool

Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is “Defeating Nondeterminism in LLM Inference” We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to…

thinkymachines's tweet image. Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is “Defeating Nondeterminism in LLM Inference”

We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to…


Aaron Wenteler reposted

Latest genomic AI models report near-perfect prediction of pathogenic variants (e.g. AUROC>0.97 for Evo2). We ran extensive independent evals and found these figures are true, but very misleading. A breakdown of our new preprint: 🧵

BrandesNadav's tweet image. Latest genomic AI models report near-perfect prediction of pathogenic variants (e.g. AUROC>0.97 for Evo2). We ran extensive independent evals and found these figures are true, but very misleading.

A breakdown of our new preprint: 🧵

Aaron Wenteler reposted

torch.manual_seed(THE_GOOD_ONE) lottery ticket hypothesis and all that

if the architecture is vanilla and the compute was cheap, you should really be questioning what makes these models so impressive for their size there are only a few possibilities 😉



Loading...

Something went wrong.


Something went wrong.