Announcing AlphaEvolve, our new LLM coding agent that has - made new scientific discoveries - discovered algorithms that are now deployed at Google (in Gemini, Transformers, TPU hardware design & data centers) Blog: deepmind.google/discover/blo... White paper: storage.googleapis.com/deepmind-med...

AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms

New AI agent evolves algorithms for math and practical applications in computing by combining the creativity of large language models with automated evaluators

deepmind.google

One of my PhD students got their visa revoked. I know of other cases amongst my AI colleagues. This is not what investing in US leadership in AI looks like. www.aljazeera.com/news/2025/4/...

US revokes nearly 1,500 student visas: Who are the targets?

Hundreds of students have had their visas cancelled and find themselves in limbo.

aljazeera.com

Congrats to UT computer scientist Swarat Chaudhuri & UT physicist Feliciano Giustino who were named as Guggenheim Fellows for 2025! #GuggFellows2025 @guggfellows.bsky.social @utaustin.bsky.social @swarat.bsky.social cns.utexas.edu/news/accolad...

Guggenheim Foundation Names 3 at UT in 100th Class of Fellows

Swarat Chaudhuri, a computer scientist, and Feliciano Giustino, a physicist, are among this year’s fellows from The University of Texas at Austin.

cns.utexas.edu

I am honored to be part of the #guggfellows2025 class. My Guggenheim project is on AI systems that can discover new math in an open-ended way. Many thanks to my students, colleagues, and mentors, who inspire me every day and without whom this work wouldn't be possible. www.gf.org/stories/anno...

Harvard has set an example for other higher-ed institutions - rejecting an unlawful and ham-handed attempt to stifle academic freedom, while taking steps to make sure students can benefit from an environment of intellectual inquiry, rigorous debate and mutual respect. Let’s hope others follow suit.

The #LeanLang Standard Library, under active development at the Lean FRO, envisions providing a reliable and extensible basis for #softwaredevelopment, #softwareverification and #mathematics through verified components, a high-quality API, performance optimization, and best-in-class documentation.

RL is so back! (well, for some of us, it never really left) awards.acm.org/about/2024-t...

Andrew Barto and Richard Sutton are the recipients of the 2024 ACM A.M. Turing Award for developing the conceptual and algorithmic foundations of reinforcement learning.

Andrew Barto and Richard Sutton as the recipients of the 2024 ACM A.M. Turing Award for developing the conceptual and algorithmic foundations of reinforcement learning. In a series of papers beginning...

awards.acm.org

Calling all scientists and students based in London! standupforscience2025.org and local groups are organizing rallies around the US to protest against the new administration’s massive and indiscriminate funding cuts to all manner of scientific research…👉🏼🧵 #sciencematters #standupforscience #london

STAND UP FOR SCIENCE

March 7, 2025. Washington DC and nationwide. Because science is for everyone.

standupforscience2025.org

Excited about Proofwala, @amitayush.bsky.social's new framework for ML-aided theorem-proving. * Paper: arxiv.org/abs/2502.04671 * Code: github.com/trishullab/p... Proofwala allows the collection of proof-step data from multiple proof assistants (Coq and Lean) and multilingual training. (1/3)

Upon learning that yesterday would be my last day as a program officer at the National Science Foundation, I shared this parting message with my colleagues. The next few months will be frenetic and stressful for them. Here are some things that you can do to help them with the mission ahead. (1)

DARPA released a Request for Information (RFI) that seeks community feedback on the draft DARPA Guide to Formal Methods to Deliver Resilient Systems for Proposals (“the FMDRS Guide”). You can find the RFI here on Sam.gov. Details in the image...

Proving the Coding Interview: A Benchmark for Formally Verified Code Generation “We introduce the Formally Verified Automated Programming Progress Standards, or FVAPPS, a benchmark of 4715 samples […] including 1083 curated and quality controlled samples” arxiv.org/abs/2502.05714

Proving the Coding Interview: A Benchmark for Formally Verified Code Generation

We introduce the Formally Verified Automated Programming Progress Standards, or FVAPPS, a benchmark of 4715 samples for writing programs and proving their correctness, the largest formal verification ...

arxiv.org

Can LLMs be used to discover interpretable models of human and animal behavior?🤔 Turns out: yes! Thrilled to share our latest preprint where we used FunSearch to automatically discover symbolic cognitive models of behavior. 1/12

This is the most relevant article to NIH and research cuts I’ve seen. Imagine if this was today , how many people would be saying “Why are we studying Gila Monsters and their impact on diabetes ? That’s wasted money !” globalnews.ca/news/9793403...

How a Canadian scientist and a venomous lizard helped pave the way for Ozempic - National | Globalnews.ca

In 1984, Dr. Daniel Drucker, an endocrinologist from the University of Toronto, discovered a hormone that helped pave the way for popular diabetes drugs such as Ozempic.

globalnews.ca

Super excited: my new @darpa program on AI for pure mathematics! Exponentiating Mathematics (expMath) aims to accelerate the rate of progress in pure math through the development of an AI collaborator and new professional-level math benchmarks. sam.gov/opp/4def3c13...

Mathematical proof assistants like Coq and Lean were made possible by a correspondence that established the equivalence between proofs and computation. Read the explainer from our archive:

The Deep Link Equating Math Proofs and Computer Programs | Quanta Magazine

Mathematical logic and the code of computer programs are, in an exact way, mirror images of each other.

buff.ly

New: The largest medical A.I. randomized controlled trial yet performed, enrolling >100,000 women undergoing mammography screening The use of AI led to 29% higher detection of cancer, no increase of false positives, and reduced workload compared with radiologists w/o AI thelancet.com/journals/lan...

Screening performance and characteristics of breast cancer detected in the Mammography Screening with Artificial Intelligence trial (MASAI): a randomised, controlled, parallel-group, non-inferiority, ...

The findings suggest that AI contributes to the early detection of clinically relevant breast cancer and reduces screen-reading workload without increasing false positives.

thelancet.com

This is one more, and such a profound, way of distinguishing between science and technology: "Technology shouts for itself; science [does not]." (And these days, some technologies truly do themselves shout…)

Mukund Thattai@thattai.bsky.social · 2y ago

Technology shouts for itself; science is more abstract and subtle. We cannot assume that the value of science is self-evident to the public. In "How Science Speaks", we will hear how scientists and science writers communicate complex ideas about the history and practice of science to non-experts.

2025 will be #mathsky interesting year!

@ayushkhaitan.bluesky.social, Amitayush Thakur, and I are organizing an #AI4Math panel at the Joint Mathematics Meeting this month. Please spread the word among your math friends! We will post a summary of the discussion after the event.

Ayush Khaitan@ayushkhaitan.bsky.social · 2y ago

Looking forward to the #jmm2025 panel on the "Use of AI tools for Mathematics research" that we are co-organizing with @swarat.bsky.social and Amitayush Thakur. The panelists are Alex Kontorovich, Rishi Mehta, Emily Wenger and Kaiyu Yang. See you there!

I really enjoy NASA Administrator (!!!) Michael Griffin on the "Real Reasons" versus the "Acceptable Reasons" to go to the moon: spaceref.com/status-repor...

An excellent post by Kevin Buzzard on informal reasoning methods like o3. The key point, one I wholeheartedly agree with, is that informal methods continue to struggle with proof even when they give the correct answers, and this is a critical liability. xenaproject.wordpress.com/2024/12/22/c...

Can AI do maths yet? Thoughts from a mathematician.

So the big news this week is that o3, OpenAI’s new language model, got 25% on FrontierMath. Let’s start by explaining what this means.

xenaproject.wordpress.com

Delighted to share our new position paper: arxiv.org/abs/2412.16075! The o1/o3 path to math reasoning is based on LLMs and large-scale test-time search. We argue for a different path that uses formal proof assistants for ✅ creating high-quality synthetic data ✅ rigorous test-time feedback. (1/2)

Good 🦋 samaritans: any of you have a friend at X who can get my X account back? If so, DM/email me! The thread below tells what happened. I couldn't ever reach a human at X; the bots kept saying X couldn't help. Every new account I've created since has been auto-suspended. bsky.app/profile/swar...

Swarat Chaudhuri@swarat.bsky.social · 2y ago

I was thinking about what to write in my first real 🦋 post. Didn't expect it to be so easy. My X account has been hacked. The hacker changed the account email, and X won't return access to me because I don't know what it was changed to. 🤡 Oh well, good riddance and sorry I didn't quit sooner.

Trustworthy AI for Science will be a big deal in <5 years. No scientist wants to have to retract their Nature paper because their AI assistant hallucinated a bad but superficially compelling result. #NeurIPS2024 #DeepThoughts #AI4Science

Yisong Yue@yisongyue.bsky.social · 2y ago

Proposing scientific theories is rational hallucination #NeurIPS2024 #DeepThoughts #AI4Science

Samuele created a fantastic benchmark for studying Neurosymbolic Reasoning Shortcuts! Reasoning shortcuts are a common failure mode of Neurosymbolic models. 🚀 Good diagnostic tools will really help the field forward! Glad to have contributed to this.

Samuele Bortolotti@samubortolotti.bsky.social · 2y ago

📣 Does your model learn high-quality #concepts, or does it learn a #shortcut? Test it with our #NeurIPS2024 dataset & benchmark track paper! rsbench: A Neuro-Symbolic Benchmark Suite for Concept Quality and Reasoning Shortcuts What's the deal with rsbench? 🧵

Yeming will present his new work on diverse generation with foundation models on 12/12 at #NeurIPS2024: openreview.net/pdf?id=sp8wH.... The main idea is to leverage synthetic data. We use influence functions to partition the data into subsets, then train multiple model adaptations on these subsets.

Arya and I'll be at #NeurIPS presenting LaSR (trishullab.github.io/lasr-web/) on Wednesday morning 11AM PST to 2PM PST (East Exhibit Hall A-C #4003). Drop by and say Hi!

If you are interested in program synthesis, you should consider attending our Neurips tutorial, where we will discuss program synthesis in the context of reinforcement learning. bsky.app/profile/prog...