Manoel Horta Ribeiro

🚨SciConBench accepted at #NeurIPS2026! Paper: arxiv.org/abs/2606.11337 Can AI *synthesize* scientific conclusions in health? We find that, on medical conclusions, we are far from it! Check out our live dashboard: sciconbench.cs.princeton.edu led by @hayoungjung.bsky.social

Can AI Agents Synthesize Scientific Conclusions? arxiv.org

🚨SciConBench accepted at #NeurIPS2026! Paper: arxiv.org/abs/2606.11337 Can AI *synthesize* scientific conclusions in health? We find that, on medical conclusions, we are far from it! Check out our live dashboard: sciconbench.cs.princeton.edu led by @hayoungjung.bsky.social

SciConBench sciconbench.cs.princeton.edu

🚨New preprint!🚨 arxiv.org/abs/2606.03095 In a randomized field experiment in a 300-level machine learning course, teaching assistants received AI-assisted feedback drafts after grading student submissions. Did AI help them provide more feedback when feedback was optional? A th…

AI Assistance for Discretionary Work: Increasing Feedback Provision in Higher Education arxiv.org

George Pólya famously said that `it is better to solve one problem five different ways than five problems one way.' He'd be sad to learn what we found: RLVR pushes LLMs in exactly the opposite direction, toward solving each problem in just one way! 📜: arxiv.org/abs/2610.11064

Measuring and Mitigating Solution Mode Collapse in RLVR arxiv.org

Paper: arxiv.org/pdf/2606.11337 Code: github.com/hayoungjungg... Data: huggingface.co/datasets/hay...

arxiv.org

What can other communities learn from WikiVault? We discuss how communities can navigate the uncertainties of adopting LLMs, learn from their use, and adapt governance along the way. Read more about these lessons and open questions in our preprint! 📄 arxiv.org/abs/2610.07660

When the Commons Appropriates a Large Language Model: How WikiVault Reshaped Korean Wikipedia arxiv.org

Paper: arxiv.org/pdf/2606.11337 Code: github.com/hayoungjungg... Data: huggingface.co/datasets/hay...

hayoungjung/SciConBench · Datasets at Hugging Face huggingface.co

New blog post! I argue that the “stochastic parrots” metaphor dismisses LLM capabilities by reducing them to “haphazardly stitching together” text... ...but retreats to the trivially true claim that LLMs predict the next token when challenged. doomscrollingbabel.manoel.xyz/p/o…

Of Mottes, Baileys, and Stochastic Parrots 🦜 doomscrollingbabel.manoel.xyz

❓❓Should LLMs have access to ACM’s Digital Library❓❓ I wrote a short essay arguing that "yes!" I try to argue that, even if you are concerned about their negative impact on science, denying them access is a poor way to address their harm! doomscrollingbabel.manoel.xyz/p/scienc…

Science Should Be Open, For LLMs Too doomscrollingbabel.manoel.xyz

Hot take: "AI slop" is a bad term and blurs the debate on the value of AI-generated content. doomscrollingbabel.manoel.xyz/p/ai-slop-isnt

AI Slop Isn’t doomscrollingbabel.manoel.xyz

New preprint! We introduce a new benchmark, SciConBench, with 9.11k scientific questions derived from Cochrane Systematic Reviews. We find evidence that frontier AI agents **cannot** synthesize scientific conclusions well. A thread 🧵 w/ @hayoungjung.bsky.social & others!

Wikipedia helped make LLMs possible. Now, it is grappling with how to use them. Unlike its peers, Korean Wikipedia embraced LLMs through WikiVault: only 0.1% of editors use it, but it accounts for ~30% of new bytes! New preprint led by @inhwa.bsky.social (who's at @acm-cscw.bsky.social!!) 🧵

@mariannealq.bsky.social and @isabelcorpus.bsky.social presenting their work on local variations on AI overviews! They find that AI overviews are much better in high information environments.