Quantum Entanglement for babies? Finally, a way to explain why they won’t nap - because their sleep state is entangled with playtime.

Being able to interpret an #ML model’s hidden representations is key to understanding its behavior. Today we introduce Patchscopes, an approach that trains #LLMs to provide natural language explanations of their own hidden representations. Learn more → goo.gle/4aS5epd

Mantis was evaluated on the NLVR^2, Birds-to-Words, QBench, our newly curated Mentis-eval and a recent dataset Mementos x.com/furongh/status…

“What would an AI research & education mission that crossed the whole campus look like?”

This is the question posed to me & a working group by
Univ. of Maryland's Provost Rice summer'23.

Now we’re announcing the answer: #AIMaryland , the AI Interdisciplinary Institute at Maryland.
>

.Univ. of Maryland launches an institute focused on ethical AI development. It supports interdisciplinary research and prepares the future workforce for an AI-infused world. CS Professor Hal Daumé III is its inaugural director.

Read more:
go.umd.edu/AIM

🤔Do you use mini-PointNet for your local point cloud geometry encoding?
😫Have you reluctantly downsampled your local point cloud due to computational constraint?
💡Try our ultra-efficent local geometry encoder VecKM! All neighborhoods are considered!
👉arxiv.org/abs/2404.01568

TACO is sort of using action-sequence state prediction (forward dynamics prediction) via a contrastive loss, although not directly resorting to inverse dynamics. arxiv.org/abs/2306.13229

Join us in shaping the future of AI!

AI Frontiers is a lab inside Microsoft Research with a mission to unlock new AI capabilities and solve real-world problems. We are hiring researchers and engineers in our teams based in Redmond and NYC.

Apply here: microsoft.com/en-us/research…

Exploring the synergy between #CausalInference and #GenerativeLLMs ! Our latest survey dives into how understanding causality can elevate LLMs' reasoning, fairness, and explainability. Plus, how LLMs themselves can push forward causal discovery. #AI #NLP #AGI #LLM #LMM

Do Large Language Models (LLM) have leadership? Together with cognitive scientists Tom Griffiths Griffiths Computational Cognitive Science Lab and Natalia Velez Natalia Vélez (natvelali @ 🧵/🟦/fediscience.org) at Princeton University, we created a multi-LLM-agent system to test if agents work better with a leader. It turns out that LLMs do have leadership,…

Happy π Day everybody! 🥧🎉 May your day be as irrational and never-ending as π itself!

Wow, check out this talk by Souradip Chakraborty on #alignment !

I totally agree. RL doesn’t have to be a method, it can just be a setting. Alignment also doesn’t have to be a narrow set of methods. It should be a problem setting.

|￣￣￣￣￣ ￣|
| This can |
| hack AI |
| now. |
| ＿＿＿＿＿_ |
(\__/) ||
(•ㅅ•) ||
/ 　 づ

Paper showing that ASCII art can get around AI guardrails. Its the return of 1980s hackers. arxiv.org/pdf/2402.11753…

Alvaro Velasquez—a program manager DARPA who leads projects on neuro-symbolic and adversarial AI—is visiting Univ. of Maryland as a guest speaker on Wednesday, March 6. The hybrid talk is co-hosted by the Washington Chapter IEEE and UMD Center for Machine Learning.

Learn more: go.umd.edu/Velasquez

Very cool work. Tokenization is indeed crucial. Curious if folks know good tokenizations for graphs.