Jack Valmadre retweeted
An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4
Jack Valmadre retweeted
Our customer just deployed what we believe is the longest-horizon, fully-automated, mobile-manipulation workflow in production.
12 linked subtasks. Running the night shift at Rectify, a London plastics manufacturer. Built and trained by the customer, using Reimagine.
1/4
Jack Valmadre retweeted
"That looks easy. Can we try?"
Our customer asked us this a few weeks into our first pilot. After a decade spent trying to crack robot deployment, that question was music to our ears. We learned the hard way that robotics must be user-first, not model-first.
Jack Valmadre retweeted
Hello world, we’re here! While today’s LLMs are primarily text prediction machines that are unreliable and have short attention spans, Metacognition is building a new architecture, which includes memory that persists, reasoning that adapts, and trust you can inspect.
Read more:
forbes.com.au/news/entrepren…
#DeepTechWeek #AI #AustralianTech #Metacognition
Jack Valmadre retweeted
AI has now solved a major open problem -- one of the best known Erdos problems called the unit distance problem, one of Erdos's favourite questions and one that many mathematicians had tried.
openai.com/index/model-dispr…
Jack Valmadre retweeted
Congratulations to @HarmonicMath on their $1.45B valuation! Their approach of requiring AI to "output reasoning as [Lean] code instead of reasoning as English" is exactly the type of effort the @VentureBeat highlighted this week when it stated "Lean4 adds a layer of trust on top of AI outputs."
Harmonic new valuation: reuters.com/business/robinho…
VentureBeat article: venturebeat.com/ai/lean4-how…
Jack Valmadre retweeted
We are proud to be Releasing NuminaMath-LEAN, a large-scale dataset of 100K mathematical competition problems formalized in Lean 4, with more than 20K human annotations.
Paired with Kimina-Prover, Kimina-autoformalizer, CombiBench, we hope these data and models will advance open-source AI for formal mathematics !
This year's IMO was bittersweet for Numina as our theorem prover did not solve any problem, we believed we were very close.
Numina is starting a new chapter after IMO 2025. While we will still continue to build SOTA open source theorem prover for high school problems, we would also like to branch out and explore undergraduate level math or even research level, open mathematical problems.
#AI4Math #FormalMath #LeanProver #AutomatedReasoning #TheoremProving #ProofAssistant #MachineLearning #AIResearch #OpenScience #Lean4 #LeanLang
Jack Valmadre retweeted
PS People seem to be quite excited that now LLMs are doing this on their own, without being “supercharged” by Lean. I’m not. The great thing about LLMs is the large quantity of text they can produce, and the speed with which they produce it. But my goal has always been pure mathematics, and understanding.
I see formalization as a tool to help ease some of the technical (boring but important) challenges of proving theorems (and sometimes the fireworks are all in the technicalities!..). But formalization is itself currently very time consuming and labor intensive - that’s where AI can come in. So for my money, formalization is in service of mathematics, and AI is in service of formalization.
I don’t care if an LLM can spit out a hundred papers a day, even if they’re mostly right; if they’re not formalized, I’m not going to bother looking at them, trying to sort through which are right or not. So while others see LLMs reasoning without formalization as a “win” (and it is one, for LLM technology), I’m personally not as interested, since it doesn’t directly serve my ultimate goal.
Two years ago: unthinkable that AI could solve a single IMO problem.
Last year: DeepMind’s AlphaProof(+ AlphaGeometry) get silver
This year: Everybody’s getting gold
Next year: the IMO is no longer a useful benchmark for AI systems, and they leave the kids alone; it’s back to just being a hard high school math test.
wsj.com/tech/ai/imo-gold-mat…
Jack Valmadre retweeted
Woohoo!! The "Medium" strength Prime Number Theorem was just proved in @leanprover Lean: (the bottom node in the picture is Green)
The main `MediumPNT` file is about 8000 lines of code, which uses a big `ZetaBounds` file with around 4000 lines of code, and another ~1000 lines for residue calculus on rectangles.
It was remarkably fun to collaborate with dozens of people, the vast majority of whom I've never met (and likely will never meet; some are math students, others are, e.g., software engineers doing some Mathlib just for fun on the weekends...), and knowing with *certainty* that their contributions are solving the tasks I set out. Quite a model for large scale math projects!
Jack Valmadre retweeted
Terence Tao has released a new #LeanLang project that connects #FormalVerification with #MathematicsEducation: The companion to "Analysis I" is intended to provide a new avenue for engaging with the proofs and exercises in Tao's foundational "Analysis I" text. 👇
Jack Valmadre retweeted
One more episode is out and this time we get foundational! Pierre-Marie Pédrot, a core Coq/Rocq dev join us in a conversation to define what even is Type Theory to begin with. What properties should we care about and why!?
typetheoryforall.com/episode…
"Oops, I Sampled It Again": Recent paper in TMLR 2024 by student Raphaël Lafargue which demonstrates that existing confidence intervals in few-shot learning benchmarks are fundamentally flawed: They sample tasks *with replacement* and then compute variance assuming independence..
🎉 Thrilled to announce our paper "Oops, I Sampled it Again: Reinterpreting Confidence Intervals in Few-Shot Learning" accepted at #TMLR2024 ! 🎉
Big thanks to my amazing coauthors, including @jvlmdr @GriponVincent ! 🙌
🔗 tinyurl.com/yc35wpnt
📽️ tinyurl.com/4k79j49m
9
673
Jack Valmadre retweeted
Joao Henriques (joao.science) and I are hiring a fully funded PhD student (UK/international) for the FAIR-Oxford program. The student will spend 50% of their time
@UniofOxford and 50% @AIatMeta (FAIR), while completing a DPhil (Oxford PhD). Deadline: 2nd of Dec AOE!!
Jack Valmadre retweeted
I'm talking at a conference later this year (on UX+AI).
I just saw an ad for the conference with my photo and was like, wait, that doesn't look right.
Is my bra showing in my profile pic and I've never noticed...? That's weird.
I open my original photo.
No bra showing.
I put the two photos side by side and I'm like WTF...
Someone edited my photo to unbutton my blouse and reveal a made-up hint of a bra or something else underneath. 🤨
Immediately, I email the conference host.
(FYI he is a great, respectable guy with 5 kids at home.)
He is super apologetic and immediately looks into the issue.
He quickly reports back that the woman running their social media used a cropped square image from their website.
She needed it to be more vertical, so she used an AI expand image tool to make the photo taller.
AI invented the bottom part of the image (in which it believed that women's shirts should be unbuttoned further, with some tension around the buttons, and revealing a little hint of something underneath). 🤯
—
FYI the conference organizers were super apologetic and took down all of the content with that photo.
Jack Valmadre retweeted
#ECCV2024 is on !
Check out our work on learning from noisy datasets where we observe that detecting out-of-distribution training images using unsupervised learning discards highly important clean examples and how to recover them.
@insight_centre @TheAIML
eccv.ecva.net/virtual/2024/p…
Jack Valmadre retweeted
✨The VOTS2024 Panel✨ will discuss the future of tracking and of VOTS. Drop your questions you'd like to see discussed in the comments below 👇.
The most intriguing🧐 question posted by 28.9. 18h CET will be rewarded a limited-edition VOTSgift🃏 ! 🎁
#VOTS2024 #AskThePanel
Jack Valmadre retweeted
The University of Adelaide and CommBank have announced a partnership to establish the CommBank Centre for Foundational Machine Learning Research.
The centre will position South Australia as a leader in foundational AI research. Click to read ➡️adelaide.edu.au/aiml/news/li…

ALT Pictured at the Centre launch are (l-r) Sabina Streatfeild, GAICD, Manager, Technology Skills Pipeline & Strategy, CBA; Professor Jessica Gallagher, Deputy Vice-Chancellor (External Engagement), University of Adelaide; Brendan Hopper, Chief Information Officer for Technology, CBA; Professor Simon Lucey, Director, AIML, University of Adelaide; The Hon. Tung Ngo MLC representing the Premier; and Professor Laura Parry, Pro Vice-Chancellor (Research Excellence), University of Adelaide.
Jack Valmadre retweeted
1/ Can AI scaling continue through 2030?
We examine whether constraints on power, chip manufacturing, training data, or data center latencies might hinder AI growth. Our analysis suggests that AI scaling can likely continue its current trend through 2030.
Jack Valmadre retweeted
kaist @kaistpr professor jinjoon "vendetta" lee @jinjoon_lee forced me to send a fairly large sum of money to his wife's bank account, under threat of withdrawing an offer to be a postdoc in his lab. university leadership such as president kwang hyung lee did nothing about this
Jack Valmadre retweeted
i was a postdoc at a korean university. one time i had plans with my girlfriend, so i told my boss i wouldn’t be coming in on saturday. he thought an appropriate response was to lock me in the lab monday (also a holiday) from 6pm to 5am, for a discussion about my work ethic




























