Binglan Li

Will your study change the world? This AI tool predicts the impact of your research A tool called Funding the Frontier visualizes all the downstream impacts of funding — and predicts which studies will have the biggest societal impact.

I have mixed feelings about this work. While it's interesting to visualize the impact of scientific works, metrics like patents and funding don't necessarily reflect social and ethical impacts, like cultural shifts.

www.nature.com/articles/d41...

A pleasant read on how Deep Learning models can facilitate the prediction of gene expression levels from DNA sequences. This question led me into the field of computational biology almost 15 years ago.

Can't wait to try out the new features of Biomni. I wonder how these will help democratize scientific data for the whole biomedical community.
biomni.stanford.edu/blog/sage-bi...
biomni.stanford.edu/blog/consens...

Why language models hallucinate OpenAI’s new research explains why language models hallucinate. The findings show how improved evaluations can enhance AI reliability, honesty, and safety.

Love how OpenAI's new paper demystifies the origin of LLM hallucination as a result of training and testing. LLMs are rewarded for making confident guesses rather than reporting uncertainty, hence returning guesses (hallucinated responses) in production.

openai.com/index/why-la...

Projects in ChatGPT | OpenAI Help Center Organize, revisit, and continue your work—all in one place

Projects are finally freely available on ChatGPT apps.

help.openai.com/en/articles/...

TransferTWAS: A transfer learning framework for cross-tissue transcriptome-wide association study TransferTWAS is a TWAS method that adaptively borrows information from multiple external tissues to boost gene-expression prediction in tissues with small sample sizes. It outperforms competing approa...

TransferTWAS: A transfer learning framework for cross-tissue transcriptome-wide association study: The American Journal of Human Genetics www.cell.com/ajhg/fulltex...

Genomics-informed drug-repurposing strategy identifies two therapeutic targets for preventing liver disease associated with metabolic dysfunction We prioritized 57 druggable targets from 212 putative causal genes of metabolic-dysfunction-associated steatotic liver disease (MASLD). Using a drug-repurposing analysis that integrated protein struct...

Genomics-informed drug-repurposing strategy identifies two therapeutic targets for preventing liver disease associated with metabolic dysfunction: The American Journal of Human Genetics www.cell.com/ajhg/fulltex...

CodeMonkeys: Scaling Test-Time Compute for Software Engineering Scaling test-time compute is a promising axis for improving LLM capabilities. However, test-time compute can be scaled in a variety of ways, and effectively combining different approaches remains an a...

It makes sense that parallel agents can improve performance, much like a team of experts can lead to better results. The study also discusses the relationship between costs and the number of agents.

scalingintelligence.stanford.edu/pubs/codemon...

What it looks like when an agent gets stuck for no reason.

It was tasked with creating a .gitignore file, following the GitHub template. It ended up reading a phantom .gitignore file.

Any suggestions on how to better prompt the agent? Do I need to break down the task?

The Paradox of Trust and Transparency in AI An AI is useless if you can't trust it—so how do we get there, and in a way that benefits and protects all parties involved?

My new piece on The Paradox of Trust and Transparency in AI www.extremetech.com/science/the-... #artificialintelligence #trust #transparency