Links: ESCA paper (NeurIPS 2025 Spotlight): https://t.co/1MGWjNMj6W ESCA dataset + model: https://t.co/igC1mtXrpK LASER repo (ICLR 2025): https://t.co/qfxjKzYwRr LASER demo:

✨✨Proud of my research group’s NeurIPS 2025 Spotlight paper on improving zero-shot performance of embodied AI agents using neurosymbolic representations! Come see us at the conference in San Diego, play with our live demo / model / dataset on HuggingFace, and check out our

Announcing our ✨ NeurIPS’25 Spotlight ✨ paper: ESCA: Contextualizing Embodied Agents via Scene-Graph Generation TLDR: We introduce a framework that grounds multimodal embodied agents in scene graphs, leading to more reliable perception, stronger reasoning, and better actions.

i've had mixed results in the past, but the new @grok model (grok-4-fast [beta]) is crushing really hard math problems that other models can't handle. it's amazing to me how fast it's improving.

my first technical job was a summer job for the computer division of a bank in cleveland ohio. i was 19 and into assembly, pascal, etc. couldn't wait to see what i would be assigned... === 1988 === my job is in the tape library. stand in front of a green-screen-matrix-terminal.

Announcing our NeurIPS paper: Once Upon an Input: Reasoning via Per-Instance Program Synthesis (PIPS) 📝: https://t.co/Rf8upKMk2c Why do LLMs (and LLM agents) still struggle on hard reasoning problems which should be solvable by writing and executing code? We find that the

@emollick i've spent serious efforts working on developing math problems with an unambiguous (e.g., numerical) answer that gpt-5-pro cannot solve. it is *nontrivial* to do so. it was totally different even 4-6 months ago.

It is easy to overlook but hard to overstate how big a role "test-time inference" has played in modern LLM gains in reasoning. I myself wasn't sure, so I put GPT-5 and Gemini-2.5 to a classic programming puzzle called variable shadowing ( https://t.co/3znsUTR58A): Both models in

Prospective PhD students: my research group has openings starting in Fall 2026! We focus on building principled yet practical systems for trustworthy AI. We are developing foundation models and agentic frameworks for critical domains such as cybersecurity, robotics, and

📢 The Software Engineering group at @Cornell_Bowers is growing fast -- we're now 8 PhD students strong! I’m recruiting PhD students for Fall 2026! If you are interested in the intersection of SE and AI, apply to Cornell CS and reach out! Ddl: Dec 15, 2025. RT!

We introduce Delta Activations, a lightweight yet powerful mechanism to represent AI models, with practical applications such as model selection ("find the best model for task instance X") and model merging ("combine models A and B without re-training to obtain a better model").

Swing by our poster session today at 11 if you're at ICML to learn more about speeding up neurosymbolic learning! We will be in the East Exhibition Hall A-B, # E-2003

Very much enjoyed advocating for symbolic reasoning for Trustworthy AI in my NSF CISE lecture, the recording is now available at

Congratulations to Dr. Ziyang Li (@_ziyang_) on defending his dissertation today! Titled "Neurosymbolic Programming in Scallop: Design, Implementation, and Applications", this dissertation proposed Scallop, a unified programming system for combining the otherwise complementary

Looking forward to discuss the promise of neurosymbolic approaches to trustworthy AI at @NSF CISE

One of the most effective things the U.S. or any other nation can do to ensure its competitiveness in AI is to welcome high-skilled immigration and international students who have the potential to become high-skilled. For centuries, the U.S. has welcomed immigrants, and this

Foundation models can now perform many reasoning tasks via prompting alone. So do we still need to train neuro-symbolic systems? Our position paper argues that neuro-symbolic prompting, not training, is the path to generalizable and interpretable reasoning.

I am alarmed by the proposed cuts to U.S. funding for basic research, and the impact this would have for U.S. competitiveness in AI and other areas. Funding research that is openly shared benefits the whole world, but the nation it benefits most is the one where the research is

We are excited to share Dolphin, a programmable framework for scalable neurosymbolic learning, to appear at ICML 2025! Links to paper and code in thread below 👇

We can’t thank @awscloud enough for the support! We are excited to see the developments in our students research!