Recently, IDT Scientist Adam Clore helped detect AI-designed protein sequences that could be biosecurity threats. The results of the study, led by @microsoft.com CSO @erichorvitz.bsky.social are now published in @science.org. 💻Read more in @technologyreview.com: idtb.io/3bfxmf

Paraphrasing a protein is like paraphrasing a sentence—the words change, but the meaning holds. Microsoft scientists applied this concept to toxic proteins, revealing a blind spot in DNA screening and leading a cross-sector team to develop a fix. msft.it/6010s4tW8

We enjoyed working with Adam Clore and his colleagues at @idtdna.bsky.social

Integrated DNA Technologies@idtdna.bsky.social · 10mo ago

Recently, IDT Scientist Adam Clore helped detect AI-designed protein sequences that could be biosecurity threats. The results of the study, led by @microsoft.com CSO @erichorvitz.bsky.social are now published in @science.org. 💻Read more in @technologyreview.com: idtb.io/3bfxmf

Call for moonshots. First phase: abstracts due on Oct 25, 2025. moonshots.laude.org/call-for-pro...

Call for Proposals // Laude Moonshots

Laude Moonshots support computer science researchers taking big, urgent swings in AI

moonshots.laude.org

How do we decide what not to share? I explore “information hazards” in this blog--on when openness in science risks misuse. Paraphrase effort led to a first-of-its-kind publishing model, accepted by Science: "When AI Meets Biology" aka.ms/AAy56z3 Full paper on the project avail here: aka.ms/AAy5e9c

Background and resources on the Paraphrase Project, published in Science last week, with some of the behind-the-scenes story, my blog on the disclosure dilemma, and a conversation with Bruce Wittman, Tessa Alexanian, and James Diggans. @msftresearch.bsky.social Full paper at: aka.ms/AAy5e9c

Microsoft Research@msftresearch.bsky.social · 10mo ago

Microsoft researchers reveal a confidential research effort that explored how open-source AI tools could be used to bypass biosecurity checks—and helped create fixes now influencing global standards. msft.it/6015sxXo7

Federal research funding works. It’s not an expense–it’s an investment. It’s not overhead–it’s a down payment on the future. - Eric Horvitz, Margaret Martonosi, Moshe Y. Vardi, and James Larus in CACM cacm.acm.org/opinion/keep... @erichorvitz.bsky.social

Study on clinician-AI collaboration: "As LLMs now demonstrate expert-level diagnostic performance, the focus shifts from whether AI can offer valuable suggestions to how it can be effectively integrated into the diagnostic workflow..." Many opportunities ahead. www.medrxiv.org/content/10.1...

I’m looking forward to the 2025 RAISE Health Symposium at Stanford University, hosted by @stanfordmedicine.bsky.social and the @stanfordhai.bsky.social. Symposium explores the today's health AI, what’s coming next, and directions forward. More here: med.stanford.edu/raisehealth/...

Conversation with the Computer History Museum on the trajectory of AI through time, on where we are today, and on directions heading into the future. www.youtube.com/watch?v=ddjN... @computerhistory.bsky.social

Chatbots Decoded Interview: Eric Horvitz

YouTube video by Computer History Museum

youtube.com

Code avail for paper, "Improving Instruction-Following in Language Models through Activation Steering" at #ICLR2025. Exciting to see influence of computed vectors across models. Paper: arxiv.org/abs/2410.12877 Repo: github.com/microsoft/ll... @alestolfo.bsky.social @besmiranushi.bsky.social

Improving Instruction-Following in Language Models through Activation Steering

The ability to follow instructions is crucial for numerous real-world applications of language models. In pursuit of deeper insights and more powerful capabilities, we derive instruction-specific vect...

arxiv.org

Alessandro Stolfo@alestolfo.bsky.social · last yr.

Our paper "Improving Instruction-Following in Language Models through Activation Steering” has been accepted to #ICLR2025! We're also excited to share that our public GitHub repo is now live. Code: github.com/microsoft/ll... Camera-ready: arxiv.org/abs/2410.12877

The 2025 AI Index report has been published. Summary & link to full report here: hai.stanford.edu/news/ai-inde... @stanfordhai.bsky.social #AI100

AI Index 2025: State of AI in 10 Charts | Stanford HAI

Small models get better, regulation moves to the states, and more.

hai.stanford.edu

www.mccormick.northwestern.edu/news/article...

What AI Means for the Future of Work

On February 18, the Center for Human-Computer Interaction + Design hosted a thought leader panel to discuss the National Academies of Sciences’s

mccormick.northwestern.edu

Seq2Symm is out now in @naturecomms.bsky.social! PLMs continue to be unreasonably efficient-- even for highly divergent sequences. Meghana and Minkyung did a great job leading on this! 📄: www.nature.com/articles/s41... 💻: github.com/microsoft/se...

Rapid and accurate prediction of protein homo-oligomer symmetry using Seq2Symm - Nature Communications

Seq2Symm is a sequence-based AI model designed to predict the symmetry of protein structures composed of identical chains, known as homo-oligomers. It surpasses existing methods in accuracy and effici...

nature.com

Just released: Report of the 2025 AAAI Presidential Panel on the Future of AI Research. Covers a selection of topics of a larger set of research directions, providing high-level glimpses. Enjoyed collaborating with colleagues on the project. Report pdf avail here: aaai.org/wp-content/u...

Background on our quantum computing announcement this week on progress on developing a topologically protected qubit: Here's an open version of my LinkedIn article on the nearly two-decade--long project: aka.ms/quantum_brea...

MedHELM announced: Popular benchmarks for medical challenge problems have become saturated w/ high perf. of LLMs. New set of benchmarks for evaluating performance LLMs in medicine. 31 datasets, clinically relevant tasks covering 22 subcategories. Leaderboard: lnkd.in/gZ7CkNqc Blog: lnkd.in/gsVQUc2P

Meeting at Asilomar 50 yrs after 1975 meeting. New issues today with fast-paced advances of AI & biology--and timeless pursuit of questions & directions on the responsibilities of scientific community. Important conversations in a beautiful setting. NPR coverage: www.npr.org/2025/02/25/n...

Great way to catch up with colleagues on the latest innovations in AI: Poster sessions! At annual meeting of the Association for the Advancement of Artificial Intelligence (AAAI) in Philadelphia.

When science & technology transcend politics: At Senate Committee hearing, Senators Maria Cantwell & Ted Cruz joined hands to showcase Microsoft’s new quantum chip that leverages discoveries at the frontiers of physics. A striking bipartisan moment. More on quantum effort: aka.ms/quantum_brea...

Top of the the year reflection. www.politico.com/sponsored/20...

Sponsored Content: Harnessing the Power and Potential of AI

How the U.S. can unlock the promise of AI while balancing innovation and responsibility

politico.com

Recent results on performance of 01-preview model on medical diagnosis & management challenges arxiv.org/abs/2412.10849 For additional reflections, see article on LinkedIn: www.linkedin.com/pulse/advanc...

Superhuman performance of a large language model on the reasoning tasks of a physician

Performance of large language models (LLMs) on medical tasks has traditionally been evaluated using multiple choice question benchmarks. However, such benchmarks are highly constrained, saturated with...

arxiv.org

Surprising scene: Jug of spilled milk in front of exhibition hall at NeurIPS meeting. Great catching up with John Platt as the anomalous event unfolded. We both had our assistants at the ready.

Super excited to preprint our work on developing a Biomolecular Emulator (BioEmu): Scalable emulation of protein equilibrium ensembles with generative deep learning from @msftresearch.bsky.social ch AI for Science. www.biorxiv.org/content/10.1...

Perspective article in Cell, "How to build the virtual cell with artificial intelligence: Priorities and opportunities." Enjoyed the collaboration with fabulous colleagues on this perspective piece: www.cell.com/cell/fulltex...

How to build the virtual cell with artificial intelligence: Priorities and opportunities

Advances in AI and omics enable the creation of AI virtual cells (AIVCs)—multi-scale, multimodal neural network models that simulate molecules, cells, and tissues across diverse states. This vision outlines their design and collaborative development, promising to transform biological research through high-fidelity simulations, accelerating discoveries, and fostering interdisciplinary open science collaborations.

cell.com

Identifying & mitigating an AI-enabled biosecurity vulnerability. Confidential effort on biosecurity red-teaming leads to distrib. of "patches" that bolster DNA synthesis screening. Article: aka.ms/AAtoa8e Work continues. @msftresearch.bsky.social

Toward AI-Resilient Screening of Nucleic Acid Synthesis Orders: Process, Results, and Recommendations

Fast-moving advances in AI-assisted protein engineering are enabling breakthroughs in the life sciences that promise numerous beneficial applications. At the same time, these new capabilities are crea...

aka.ms

Processes for seeking mutual understanding or “grounding,” adapted from psychology, are important in designs for agentic AI systems, per their complexity & potential high stakes of actions taken on behalf of users. New position paper with colleagues: tinyurl.com/mr2jr2rx @msftresearch.bsky.social

The Pritzker Prize: Honoring groundbreaking contributions at the intersection of AI and the natural sciences and engineering. Learn more and nominate here: datascience.uchicago.edu/research/ai-...

Co-organizers of inaugural AI-Bioscience Collaborative summit at National Academies. Event brought together int'l leaders in AI & biotechnology to explore bolstering sequence datasets, compute tools, resources, and to discuss key topics, including AI & biosecurity. More here: tinyurl.com/y6nwa5ze

Bringing a couple of recent thought pieces to BlueSky. What do folks think about these top 10 priorities for AI research, policy, and practice? A recent CACM article from the leads and cofounder of the One Hundred Year Study on AI. cacm.acm.org/opinion/now-...