Being presented Wednesday, July 30 11:00-12:30 in Hall 4/5. Also featured at GEM (Generation, Evaluation, & Metrics) Workshop on Thursday July 31st. Two chances to learn about LLMs and finance!. (Work done with the GT Fintech lab; I am not an author).

ArXiv: GitHub: HuggingFace:

At #ACL2025? Wondering if LLMs can do finance tasks? . My PhD student @GlennMatlin will be presenting the Finance Language Model Evaluation (FLaME) benchmark task.

Please consider applying if you are a Doctoral student working in the broad areas of responsible computing, responsible AI, human-centered AI, human-AI interaction, pluralistic value alignment, policy, or any related area.

Applications are open for the Doctoral Consortium at the 2025 Summit on Responsible Computing, AI, and Society Taking place October 27, 2025. Applications received by September 15 will receive full consideration.

RT @Thinkwert: Prescient, from Shel Silverstein in 1981

You should submit a contribution and/or attend the Summit on Responsible Computing, AI, and Society. The call is for contribution is here: . We seek to surface key visions for AI, human-AI interaction, human-centered/pluralistic value alignment & similar.

Dr. Chowdhury is a pioneer in the responsible, ethical, explainable and transparent AI. She is the first person to be appointed by the Department of State as the US Science Envoy for AI. She co-developed the first-in-industry algorithmic tool to mitigate bias in AI systems.

Hi all. I am delighted to announce our first Keynote speaker for the 2025 Summit on Responsible, Computing, AI, and Society: Dr. Rumman Chowdhury

RT @tourzhao:

Who wants a $20M supercomputer for public AI projects? Georgia Tech is leading the effort.

There should be a Philip K Dick-verse where all the PKD stories exist in the same universe and the protagonists from the different stories occasionally come together to solve major existential crises.

RT @danfei_xu: Honored to receive the NSF CAREER Award from the Foundational Research in Robotics (FRR) program!. Deep gratitude to my @ICa….

"Novelty detection" is the first step in developing safe RL systems that can shut themselves down or adapt online. Dynamic online adaptation is hard. Geigh has some upcoming work on that. Ask him about it.

Importantly, our technique does not require setting a threshold hyperparameter for how much distribution shift is allowable. Prior techniques in the RL setting required such a threshold, making assumptions about what can be known in advance about dynamics changes.

As RL agents are deployed in the real world, they will face situations that they have never trained for because the real world is complex and messy. Moreover in the lab we tend to forget that transition dynamics can change. If trained with a world model, we measure surprise.

Keep an eye out for Geigh Zollicoffer at #ICML25 His Spotlight Poster is on using world models to detect when environment transition dynamics have changed and are out distribution for an RL agent

Not everything made it into this short news story, but if you are interested, the longer version is is essentially covered in this BlueSky thread:

I talked with CNN about what the hell is going on with Grok

RT @UpolEhsan: 🎉 Energized to see this coverage from my academic home, @Northeastern, on my @Harvard 𝐁𝐞𝐫𝐤𝐦𝐚𝐧 𝐊𝐥𝐞𝐢𝐧 𝐅𝐞𝐥𝐥𝐨𝐰𝐬𝐡𝐢𝐩, which explor….