🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

We suspect OpenAI's books2 dataset might be "all of libgen", but no one knows. It's all pure conjecture. Nonetheless, books3, released above, is "all of bibliotik", which I imagine will be of interest to anyone doing NLP work. Or anyone who wants to read 196,640 books. :)

🧬💻👾 celebrating a year of great achievements with old (Roberto Sirica ), current (Silvia Buonaiuto Flavia Villani Gianluca Damaggio Andrea Guarracino), remote (Maddie, Franco), favorite (Erik Garrison) collaborators🍾🥇🎊 looking forward for next year's projects CNR IGB - Institute of Genetics and Biophysics ABT and The University of Tennessee Health Science Center 🦸🦇❤️

📢Introducing In-Context Vectors (ICV), a more effective + controllable alternative to #LLM in-context learning arxiv.org/abs/2311.06668 Instead of prompting, we learn LLM latent vector that captures user examples. Then we can steer text generation w/ ICV to get better results

Finally, one of the cool insights that came from this work is that dark matter in biology (missing heritability) is arguably (via the equivalence of the Price equation and the virial theorem) an actual analog of dark matter in astronomy! 15/15 technologyreview.com/2010/12/21/197…

🧬 Do you think pangenomics is hot? Attend our Workshop, Conference & Biohackathon (May 18-22, 2024). Discover genomic diversity, engage with leading scientists, and contribute to cutting-edge software. Register now! pangenome.github.io/MemPanG24/ #Pangenomics #Bioinformatics #MemPanG24

I have spent the last few months writing a simple transformer program that can be used easily on pangenome data. 1/n GitHub - mol-evol/panGPT: A Transformer for Pangenome data github.com/mol-evol/panGPT

For almost a year I've been using whisper speech to text instead of typing github.com/ekg/loq

Three/four exciting 4-year computational post-doctoral positions in our team Human Technopole to work on a variety of projects. Please retweet and share within your networks.

One project develops computational approaches to infer human variation from graph genomes and its role in driving human phenotypic variation Davide Bolognini @raveancic Erik Garrison Peter Sudmant. See also biorxiv.org/content/10.110…. Apply here: careers.humantechnopole.it/o/postdoc-in-p…

Our paper on the amylase locus is out in Nature (doi.org/10.1038/s41586…)! Working with this team was incredibly fun! We developed a computational framework (cosigt) to infer the haplotype composition of short read samples leveraging the pangenome graph. 1/n

In a new preprint, Ragnar {Groot Koerkamp} 🦋 and I led the derivation of a new lower bound, g’ (red), on k-mer sampling scheme density: - our bound appears to be tight for k ≡ 1 (mod w). - existing schemes are nearly optimal! - mod-minimizers are optimal for large sigma for k ≡ 1 (mod w)

making a lot of pairwise alignments (PAF) and needed a way to confirm validity of both coordinates and alignment github.com/ekg/pafcheck

cool project where researchers applied interpretability techniques that have worked on language models to a model that uses protein sequences to predict protein structures. And indeed, the models are discovering real biological concepts that humans have also found.

Memory makes computation universal, remember? thinks.lol/2025/01/what-c…

The plan at FutureHouse has been to build scientific agents and use them to make novel discoveries. We’ve spent the last year researching the best way to make agents. We’ve made a ton of progress and now we’ve engineered them to be used at scale, by anyone. Today, we’re launching

Postdoc position opening in my group! Research projects: pangenomes for diverse organisms, genome evolution, biocomputing, language models. Please reach out if interested!