Author | Lightnews

Sebastien Bubeck @sbubeck.bsky.social · Aug 7

www.youtube.com/live/0Uu_VJe...

10am PT today!

Introducing GPT-5

YouTube video by OpenAI

www.youtube.com

2 6

Sebastien Bubeck @sbubeck.bsky.social · Jun 12

This was a really fun conversation with @billgates.bsky.social and Peter Lee, and I hope you enjoy it too!

(roughly we talk about our respective first contacts with genAI/LLM/sparks and what we see as missing steps/future steps for it to truly transform healthcare)

youtu.be/N0w9uO7kH80?...

How AI is reshaping the future of healthcare and medical research

YouTube video by Microsoft Research

youtu.be

1 6

Reposted by Sebastien Bubeck

The Onion @theonion.com · May 2

Impressive ‘GTA VI’ Trailer Features Characters Claiming They’re Sentient, Begging For Release From Digital Prison
theonion.com/impressive-g...

38 320 3.3K

Sebastien Bubeck @sbubeck.bsky.social · Mar 25

8

Sebastien Bubeck @sbubeck.bsky.social · Feb 2

Correct!

3

Sebastien Bubeck @sbubeck.bsky.social · Jan 31

Overall I think o3-mini will be a very useful model for the academic community.

Learn more here: openai.com/index/openai...

OpenAI o3-mini

Pushing the frontier of cost-effective reasoning.

openai.com

5

Sebastien Bubeck @sbubeck.bsky.social · Jan 31

Interestingly though the reference it gives is not quite the correct one, but it is very closely related! In general I have found that references are "fuzzily correct", giving some mixed up version of authors/journals/titles, but surprisingly still useful!

1 6

Sebastien Bubeck @sbubeck.bsky.social · Jan 31

o3-mini is a remarkable model. Somehow it has *grokked arxiv* in a way that no other model on the planet has, turning it into a valuable research partner!

Below is a deceitfully simple question that confuses *all* other models but where o3-mini gives an extremely useful answer!

2 15

Sebastien Bubeck @sbubeck.bsky.social · Jan 8

With MIT license.

Enjoy!

huggingface.co/microsoft/ph...

microsoft/phi-4 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

3 6 27

Sebastien Bubeck @sbubeck.bsky.social · Jan 3

Nice to see SLMs making it to the 10 Breakthrough Technologies of last year!!!

www.technologyreview.com/2025/01/03/1...

10 Breakthrough Technologies 2025

What will really matter in the long run? That’s the question we tackle each year as we compile this annual list.

www.technologyreview.com

2 3 11

Sebastien Bubeck @sbubeck.bsky.social · Dec 13

and if you're the type to be worried about contamination (American Mathematics Competition from last month, after phi-4 was trained):

2 11

Sebastien Bubeck @sbubeck.bsky.social · Dec 13

1 1 5

Sebastien Bubeck @sbubeck.bsky.social · Dec 13

arxiv.org/abs/2412.08905 Hope you like it!

Phi-4 Technical Report

We present phi-4, a 14-billion parameter language model developed with a training recipe that is centrally focused on data quality. Unlike most language models, where pre-training is based primarily o...

arxiv.org

3 5 38

Sebastien Bubeck @sbubeck.bsky.social · Dec 9

Pretty crazy ...

1 8

Sebastien Bubeck @sbubeck.bsky.social · Dec 6

It's getting there ...

2 15

Sebastien Bubeck @sbubeck.bsky.social · Dec 4

They made a pretty cool poster for the event too 😄

2 2 15

Sebastien Bubeck @sbubeck.bsky.social · Dec 4

Tomorrow morning at the Simons Institute : Sparks versus embers, can LLMs solve major open mathematical conjectures?

(FWIW I agree with everything in the Embers paper, so I guess the debate will be about the conclusions to draw from current evidence!)

simons.berkeley.edu/talks/sebast...

Debate: Sparks versus embers

Debaters: Sebastien Bubeck (OpenAI), Tom McCoy (Yale) Discussants: Pavel Izmailov (Anthropic), Ankur Moitra (MIT) Moderator: Anil Ananthaswamy

simons.berkeley.edu

6 20

Sebastien Bubeck @sbubeck.bsky.social · Dec 3

Sure

1

Sebastien Bubeck @sbubeck.bsky.social · Dec 2

Yes, I was a bit hyperbolic:-). One of the cleanest open problem that is still open and doable I think is to prove a n^{3/2} sqrt(T) lower bound for BCO. I'd love to see that.

1

Sebastien Bubeck @sbubeck.bsky.social · Dec 2

This should be interesting!

forum.openai.com/public/event...

Virtual Event: The Future of Math with o1 Reasoning - Event | OpenAI Forum

About the Talk:Fields Medal-winning mathematician Terence Tao makes his second appearance in the OpenAI Forum alongside OpenAI’s SVP of Research, Mark Chen to explore a future where mathematics and ar...

forum.openai.com

2 11 48

Sebastien Bubeck @sbubeck.bsky.social · Dec 1

I guess before various versions of what's described here www.scientificamerican.com/article/ai-w... take place

AI Will Become Mathematicians’ ‘Co-Pilot’

Fields Medalist Terence Tao explains how proof checkers and AI programs are dramatically changing mathematics

www.scientificamerican.com

4

Sebastien Bubeck @sbubeck.bsky.social · Dec 1

Some more of the resolutions that happened in the last few years:
arxiv.org/abs/2406.18672
arxiv.org/abs/2406.06506
arxiv.org/abs/2303.04752
arxiv.org/abs/2211.05753
arxiv.org/abs/2111.11316

Eve, Adam and the Preferential Attachment Tree

We consider the problem of finding the initial vertex (Adam) in a Barabási--Albert tree process $(\mathcal{T}(n) : n \geq 1)$ at large times. More precisely, given $ \varepsilon>0$, one wants to outpu...

arxiv.org

3 6

Sebastien Bubeck @sbubeck.bsky.social · Dec 1

It's looking like all the open problems I have thought about in the last 10 years are now solved (or in some cases on the verge of being solved)? Latest case in point this beautiful new paper: arxiv.org/abs/2411.18614 . I'm glad we (humans) got all this results just in time!

Optimal root recovery for uniform attachment trees and $d$-regular growing trees

We consider root-finding algorithms for random rooted trees grown by uniform attachment. Given an unlabeled copy of the tree and a target accuracy $\varepsilon > 0$, such an algorithm outputs a set of...

arxiv.org

3 5 31

Sebastien Bubeck @sbubeck.bsky.social · Nov 27

Was there a paper hinting at that 18 months ago? Hmmm 🤔🤣🤣

Ethan Mollick @emollick.bsky.social · Nov 27

It is funny how much of the dominant discussion about LLMs eighteen months ago (can AI pass the Turing test? can it do tasks that are not explicitly in the training data? is it a stochastic parrot?) faded. Lots of questions left (if/when it can reason, etc) but a big quiet shift in assumptions.

1 7

Reposted by Sebastien Bubeck

Thomas Wolf @thomwolf.bsky.social · Nov 22

Great AI people starter pack from @chris.bsky.social!

go.bsky.app/KRsy8pF

AI Bluesky

Join the conversation

go.bsky.app

7 12 73