Shiyi Zoe Du

Spent some time updating my Awesome-Science-Agents repo, a curated list of papers on LLM agents for scientific discovery. The pace in this space is honestly wild. A year ago “AI Scientist” felt like a bold experiment. Now we have: → Kosmos (FutureHouse) doing ~6 months of PhD-level work in a single 12-hour run → Virtual Lab (Stanford, Nature 2025) designing 92 SARS-CoV-2 nanobodies, with several validated in the wet lab → AI co-scientist (Google) replicating a decade of unpublished antimicrobial resistance research in days → MARS (Matter 2026) closing the loop with robots for autonomous materials discovery → AI Scientist-v2 (Sakana) producing the first AI-generated paper to pass peer review Also worth reading the more sobering “Why LLMs Aren’t Scientists Yet” (Jan 2026) - 3 out of 4 autonomous research attempts failed. The honeymoon is ending; the hard problems are showing up. The repo now spans ~70 papers across general science agents, benchmarks, physical sciences, life sciences, and social sciences - from 2020 to early 2026. If you’re working on something in this space - a paper, tool, benchmark, or project - I’d love to hear about it. Feel free to reach out or open a PR; I’d be happy to keep expanding the list and hopefully make it useful for more people in the community 🙌 🔗github.com/zoedsy/awesome… #AIScientists #LLM #Agents #ScientificDiscovery

We want to help scientists discover their next breakthrough with AI. Gemini for Science is our new suite of experimental tools to help them explore more hypotheses, validate work at scale, unpack literature with ease, and more 🧵

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

(1/4) Can remembering more of the past make AI agents less cooperative? In our new paper, we study LLM agents in repeated social dilemmas. The key variable is not how many rounds they play, but how much prior interaction history they can access when making each decision.

Thank you Elvis for introducing our paper about "The Memory Curse in LLM Agents"!

// The Memory Curse in LLM Agents // (bookmark it) Long histories apparently degrades agents as they become increasingly history-following and risk-minimizing. Across 7 LLMs and 4 social dilemma games over 500 rounds, expanding accessible history degraded cooperation in 18 of 28 model–game combinations. They call it the memory curse. Lexical analysis of 378,000 reasoning traces shows the mechanism: it's not that agents become paranoid, it's that forward-looking intent erodes. Long histories pull the model into reasoning about past slights instead of future payoffs. A LoRA adapter trained only on forward-looking traces mitigates the decay and transfers zero-shot to new games. Memory sanitization, keeping prompt length fixed but swapping in synthetic cooperative records, restores cooperation, proving the trigger is content, not length. And ablating explicit Chain-of-Thought often reduces the collapse, meaning deliberation actively amplifies the curse. Paper: arxiv.org/abs/2605.08060 Learn to build effective AI agents in our academy: academy.dair.ai

Congrats to TML team on the new release!

feel like research taste is, at its core, about having a sense for the future and being willing to bet on it.🤣

Thanks so much for all the interest and DMs! Also, thank you so much to everyone who reposted, liked, bookmarked, or helped share this. I received more messages than expected, so may not be able to reply to everyone immediately. I’ll try to forward relevant profiles to him, and his team may reach out if there’s a good fit (full time is preferred. Thanks again!

@lrzneedresearch 羡慕startup

@zzzoooeee321 羡慕student researcher

My friend at Google is urgently looking for Student Researchers for Research projects related to coding agents during summer. Part-time is possible. If you’re interested and have relevant experience, feel free to DM me!

⚙️ Our approach? CodonMoE is a plug-and-play adapter for pretrained DNA language models. It reshapes nucleotide-level embeddings into codon-level representations, applies expert-based codon transformations, and integrates the enriched representations back for mRNA prediction.

Excited to share that our paper “CodonMoE: DNA Language Models for Codon-Dependent mRNA Prediction” has recently been accepted by Bioinformatics! 🎉 CodonMoE is a lightweight plug-and-play adapter that enables existing DNA language models to perform mRNA analysis tasks effectively — without RNA-specific pretraining. Across tasks spanning mRNA stability, expression, and regulation, CodonMoE consistently improves DNA models while using substantially fewer parameters than specialized RNA models.🧬 📎 Paper: arxiv.org/abs/2508.04739 Big thanks to my collaborators and advisor: Litian Liang, Jiayi Li, and Carl Kingsford🙌 Comments and feedback are very welcome! #GenomicLanguageModels #AIforScience #PEFT #mRNA

🧬 Why CodonMoE? Genomic language models often require either separate DNA/RNA-specific models or large multimodal architectures. CodonMoE asks whether we can reuse pretrained DNA models for codon-dependent mRNA prediction — without RNA-specific pretraining.

Excited to share my first PhD work! We show that imperfect rewards in policy gradient are not always harmful, some can be benign or even beneficial. A lot of fun working with @hubstrauss @stanleyrwei @prfsanjeevarora and @noamrazin 😀

@zekun_jiang Thank you Zekun!!🥳

@zzzoooeee321 Good job~

Spent some time updating my Awesome-Science-Agents repo, a curated list of papers on LLM agents for scientific discovery. The pace in this space is honestly wild. A year ago “AI Scientist” felt like a bold experiment. Now we have: → Kosmos (FutureHouse) doing ~6 months of PhD-level work in a single 12-hour run → Virtual Lab (Stanford, Nature 2025) designing 92 SARS-CoV-2 nanobodies, with several validated in the wet lab → AI co-scientist (Google) replicating a decade of unpublished antimicrobial resistance research in days → MARS (Matter 2026) closing the loop with robots for autonomous materials discovery → AI Scientist-v2 (Sakana) producing the first AI-generated paper to pass peer review Also worth reading the more sobering “Why LLMs Aren’t Scientists Yet” (Jan 2026) - 3 out of 4 autonomous research attempts failed. The honeymoon is ending; the hard problems are showing up. The repo now spans ~70 papers across general science agents, benchmarks, physical sciences, life sciences, and social sciences - from 2020 to early 2026. If you’re working on something in this space - a paper, tool, benchmark, or project - I’d love to hear about it. Feel free to reach out or open a PR; I’d be happy to keep expanding the list and hopefully make it useful for more people in the community 🙌 🔗github.com/zoedsy/awesome… #AIScientists #LLM #Agents #ScientificDiscovery

typo a year ago -> many years ago