Sign in

Dan Roy

@roydanroy.bsky.social
8.5K followers 552 following 154 posts

Research Director, Founding Faculty, Canada CIFAR AI Chair @VectorInst. Full Prof @UofT - Statistics and Computer Sci. (x-appt) danroy.org I study assumption-free prediction and decision making under uncertainty, with inference emerging from optimality.

PostsRepliesMedia
Dan Roy @roydanroy.bsky.social · 21/04/2025
Tian and Karolina and team are at ICLR. Come say hi.
081
Dan Roy @roydanroy.bsky.social · 07/04/2025
Curious. Didn’t know meta had a PPL team.
020
Dan Roy @roydanroy.bsky.social · 07/04/2025
I like to think about non-reasoning model responses as vibes.
030
Dan Roy @roydanroy.bsky.social · 07/04/2025
So who’s read the 2027 article? What do you think?
410
Dan Roy @roydanroy.bsky.social · 23/03/2025
Someone has suggested I check out bsky again. So I'm back looking around here. Notification list is kinda boring. So any good conversations going on? Perhaps about LLM/AI reasoning?
8340
Dan Roy @roydanroy.bsky.social · 23/03/2025
Of course.
000
Dan Roy @roydanroy.bsky.social · 03/02/2025
Anyone else have the worry that a lot of LLM research is .... just bad psychology?
10702
Dan Roy @roydanroy.bsky.social · 03/02/2025
And, to achieve the results in this paper, what was the most challenging part? Why had previous attempts fallen short? What was your key new insight?
220
Dan Roy @roydanroy.bsky.social · 02/02/2025
Very interesting. So, what was the biggest hole to fill, in terms of hypotheses?
120
Reposted by Dan Roy
Mary Beth Hunzaker @mbhunzaker.bsky.social · 08/01/2025
Okay, so just a few* thoughts (*this got longer as I wrote 😅….long thread)-
15232
Dan Roy @roydanroy.bsky.social · 08/01/2025
Acknowledgments.
070
Dan Roy @roydanroy.bsky.social · 08/01/2025
I got to ski Revelstoke this winter break. Couple observations: the price of receiving 600 cm of snow by Jan 8 is that it is constantly snowing. Saw almost no sun the whole time and the peak was often in whiteout conditions (though North Bowl was always clear…). See image for more.
0100
Dan Roy @roydanroy.bsky.social · 08/01/2025
Multiple friends have likely lost their homes in Los Angeles. Can’t imagine how disorienting this would be. They had only minutes to flee and grab belongings.
070
Dan Roy @roydanroy.bsky.social · 30/12/2024
What are the key papers to read?
160
Dan Roy @roydanroy.bsky.social · 22/12/2024
OK. Practical question times. How are you adjusting your research given progress in reasoning style models? Also how are you adjusting the way you work?
11594
Dan Roy @roydanroy.bsky.social · 15/12/2024
A $100,000,000 experiment is no longer "consequence" free. Ilya is saying "scaling is over", but this may simply be that the scaling "laws" (not laws) are no longer accurate. Also, those laws are tied to hyperparameter tunings.
130
Dan Roy @roydanroy.bsky.social · 14/12/2024
Sure some were empirical. Some were not.
000
Dan Roy @roydanroy.bsky.social · 14/12/2024
I'd say no in a sense. Xavier-He initialization was theoretical work. And that was absolutely critical.
120
Dan Roy @roydanroy.bsky.social · 13/12/2024
Pretraining is not done. It's just that theorists haven't told the hackers how to do it better.
3350
Dan Roy @roydanroy.bsky.social · 13/12/2024
Annoying. If it could be automatic, sure.
100
Dan Roy @roydanroy.bsky.social · 12/12/2024
I'd say wait then.
010
Dan Roy @roydanroy.bsky.social · 12/12/2024
That's part of the spec. I don't think this is too problematic. The example they give is problems in NP, where there is a polynomial time checker (i.e., a polytime EV), but generating an instance that passes the checker is hard in the worst case.
010
Dan Roy @roydanroy.bsky.social · 12/12/2024
Now that I've had a taste of X without post length limitations, I've got to say that it is quite annoying have to fit tweets into 256 characters here on bsky. On X, when they get to long, they go below the fold, and so you're still incentivized to make it short. Can't we have that here?
9241
Dan Roy @roydanroy.bsky.social · 11/12/2024
Lottery ticket?
130
Dan Roy @roydanroy.bsky.social · 11/12/2024
@gkdziugaite.bsky.social. Works at GDM and Mila. Influential, technical work.
130
Dan Roy @roydanroy.bsky.social · 11/12/2024
OK
100
Dan Roy @roydanroy.bsky.social · 11/12/2024
Many of these sound to be very problematic if you hope that the result would be accepted by the mathematical community. E.g. "The proof appears to use computational evidence (listing out cases) as a substitute for theoretical proof." It seems you're not meeting the usual standard.
100
Dan Roy @roydanroy.bsky.social · 11/12/2024
Please ask Claude. What would likely be the chief criticisms of my argument above were I to submit it to a traditional mathematical journal.
110
Dan Roy @roydanroy.bsky.social · 10/12/2024
I've now read this paper carefully if anyone wants to discuss it.
470
Dan Roy @roydanroy.bsky.social · 06/12/2024
Great analogy.
000
Dan Roy @roydanroy.bsky.social · 06/12/2024
I'm obviously not talking about blocking bots.
100
Dan Roy @roydanroy.bsky.social · 06/12/2024
I've never understood people who block people on Twitter, outside of cases of harassment. Such weak sauce.
360
Dan Roy @roydanroy.bsky.social · 06/12/2024
Link: arxiv.org/abs/2412.024...
arxiv.org
Artificial Expert Intelligence through PAC-reasoning
Artificial Expert Intelligence (AEI) seeks to transcend the limitations of both Artificial General Intelligence (AGI) and narrow AI by integrating domain-specific expertise with critical, precise reas...
181
Dan Roy @roydanroy.bsky.social · 06/12/2024
Yes, that's it! Should have posted the link.
120
Dan Roy @roydanroy.bsky.social · 06/12/2024
Would really appreciate any insights people might be able to offer!
110
Dan Roy @roydanroy.bsky.social · 06/12/2024
What I'm curious about is whether there's a compelling case to spring for a Macbook Pro M4 or one of the Pro/Max model from the perspective of being able to do "inference" in one of the more interesting models on the laptop itself, without having to deal with an external machine/cloud, etc. [2/3]
220
Dan Roy @roydanroy.bsky.social · 06/12/2024
Has anyone out there compared the GPU/memory specs for the new Macbook Pro M4 Pro/Max computers to the requirements to run inference in the more interesting open source LLM models, especially those aimed at reasoning? [1/2]
280
Dan Roy @roydanroy.bsky.social · 06/12/2024
Has anyone read the new Shalev-Shwartz paper on PAC reasoning yet and want to discuss?
580
Dan Roy @roydanroy.bsky.social · 03/12/2024
Have you handed your image to, say, Claude? I suspect it predicts the response of professional mathematicians pretty well.
300
Dan Roy @roydanroy.bsky.social · 03/12/2024
How does this relate to the following? If you have a continuous function from a space S to 2-point discrete space, then the inverse image of {0} or {1} must be both closed and open. In most natural continuous spaces, the only clopen subsets are the empty set and the whole set S.
000
Dan Roy @roydanroy.bsky.social · 02/12/2024
Data collection.
050
Reposted by Dan Roy
Terence Tao @teorth.bsky.social · 02/12/2024
*AI for Mathematics and Theoretical Computer Science*, a workshop hosted jointly by the Simons Institute for the Theory of Computing and the Simons Laufer Mathematical Sciences Institute, will be held in Berkeley at April 7-11, 2025. simons.berkeley.edu/workshops/si...
simons.berkeley.edu
Simons Institute for the Theory of Computing and SLMath Joint Workshop: AI for Mathematics and Theoretical Computer Science
This is an exciting time for mathematics, as new technologies for mathematical reasoning provide novel opportunities for mathematical research, communication, and discovery. Mathlib, a library of form...
113027
Dan Roy @roydanroy.bsky.social · 02/12/2024
See you there.
100
Dan Roy @roydanroy.bsky.social · 02/12/2024
Going to try Hobbit and His Dark Materials. He's already read a number of Percy Jackson, Wings of Fire, How to Train Your Dragon. We read Narnia together a while ago.
280
Dan Roy @roydanroy.bsky.social · 02/12/2024
Great initiative! Thanks :)
020
Dan Roy @roydanroy.bsky.social · 02/12/2024
Huge community response to this. Much appreciated.
1170
Dan Roy @roydanroy.bsky.social · 01/12/2024
Too violent? Or the moral ambiguity of Enders' actions?
100
Dan Roy @roydanroy.bsky.social · 01/12/2024
My 9 yo has finished all the Harry Potters. Any suggestion on where to go from here? Maybe age appropriate sci fi?
50331
Dan Roy @roydanroy.bsky.social · 01/12/2024
Oh, ha. This. I remember this.
010
Reposted by Dan Roy
Daniel Bergstresser @dbergstresser.bsky.social · 29/11/2024
Poem by Joseph Fasano
10488791139