Pradeep Ravikumar
@RavikumarPrad
Followers
416
Following
19
Media
3
Statuses
85
Professor, Machine Learning @ CMU; co-Editor-in-Chief, Journal of Machine Learning Research (JMLR); Third-wave AI
Pittsburgh, PA
Joined January 2024
When you embed things, you unleash their awesome power.
3
8
20
This Foundation tv show has managed to find the lowest level of quality that remains enjoyable to me. I can acknowledge in real time that this is not very good. And yet I will watch every episode upon release.
327
42
2K
GPT-5 achieves 94.6% accuracy on AIME 2025, suggesting near-human mathematical reasoning. Yet ask it to query your database, and success rates plummet to the teens. The Spider 2.0 benchmarks reveal a yawning gap in AI capabilities. Spider 2.0 is a comprehensive text-to-SQL
6
4
43
@Noahpinion I am starting to like GPT5-Thinking more now
0
1
1
would be ironic if this was generated by an LLM that was prompted to not sound like an LLM.
0
1
8
Embrace openness.
In the current AI talent war, everyone is focused on the big numbers (alleged compensation packages). It misses the bigger picture: the cultural shift following the DeepSeek moment. META is the American leader in open science (publications) and open source (Llama). Both OpenAI
37
24
382
Your chance of being attacked by a squirrel is low, but never zero.
3K
10K
78K
📢 Announcing NeuS 2025 - 2nd International Conference on Neuro-symbolic Systems! 📢 Join us May 27-30, 2025, at the University of Pennsylvania, Philadelphia. More info: https://t.co/PbaUcQhCsS
#NeuS2025
0
5
10
Carnegie Mellon University at NeurIPS 2024 – Machine Learning Blog | ML@CMU | Carnegie Mellon University Carnegie Mellon University is proud to present 194 papers at the 38th conference on Neural Information Processing Systems (NeurIPS 2024)
blog.ml.cmu.edu
Carnegie Mellon University is proud to present 194 papers at the 38th conference on Neural Information Processing Systems (NeurIPS 2024), held from December 10-15 at the Vancouver Convention Center....
5
42
320
This was weirdly engaging, I expect we will be consuming more and more AI videos and content
0
0
2
We need a class of "simple" problems that neural models are bad at; and where simple extensions such as plonking a simple reasoner or planner on top of them are also not likely to work
Moravec's paradox in LLM evals I was reacting to this new benchmark of frontier math where LLMs only solve 2%. It was introduced because LLMs are increasingly crushing existing math benchmarks. The interesting issue is that even though by many accounts (/evals), LLMs are inching
0
0
9
Likelihood-based Differentiable Structure Learning
arxiv.org
Existing approaches to differentiable structure learning of directed acyclic graphs (DAGs) rely on strong identifiability assumptions in order to guarantee that global minimizers of the...
0
9
61
very excited to announce that our math department at CMU has an open tenure-track position at the intersection of math and ML! CMU math is a fantastic environment for research in this area with world-class ML and CS departments next door. please apply! https://t.co/BHuRPCyykc
0
10
19
Very ironically, one corollary of @binarybits beautiful point below is that rationalists do not think deep enough: it takes shallow thinking, tunnel vision, to imagine the world is like a chess board, or that scaling of transformers would yield AGI.
This is close to right but not quite. Rationalists think most problems are like chess—"deep" and not too fact-intensive—such that you can make progress just by thinking harder. Empiricists think most problems are shallow—the hard part is usually getting the right knowledge.
1
2
19
Is this overfitting or underfitting?
I have a follow-up question: what are people's favorite examples of overfitting?
0
0
3
Conditioning on the two things noted previously, I don't see it affecting the Biden aide decisions too much.
0
0
0
I still think Biden will be the Dem nominee: the Bayesian reasoning below still largely applies. Backlash from donors and other elites is probably harsher than Biden aides might have anticipated. It's worth considering that they probably anticipated some backlash.
Bayesian reasoning suggests that Biden will be the Dem nominee. We have to condition on two things: the Biden team knew his state well and how it comes across, and that the team is smart and power-savvy.
1
0
1