Samuel L Smith

279 posts

Samuel L Smith

Samuel L Smith

@SamuelMLSmith

Member of Technical Staff at OpenAI. Formerly Staff Research Scientist at Google DeepMind. Ex-Physicist.

Присоединился Ocak 2021
387 Подписки3.8K Подписчики
Закреплённый твит
Samuel L Smith
Samuel L Smith@SamuelMLSmith·
The Training team @OpenAI is hiring researchers in London 🚀 Our twin missions are to train better LLMs, and serve them more cheaply Get in touch if you are excited to collaborate on architecture design, reliable scaling, and faster optimization
English
11
38
490
88.7K
Apuma
Apuma@CoachApumaYTube·
@SamuelMLSmith @OpenAI @SamuelMLSmith Hi sam. Could you pls clarify: Is this latency improvement the result of simply reducing Juice / Thinking tokens of Gpt5.2 models? Or is this 40% Entirely from inference stack optimisations? Everyone is wondering about this, as the prior impacts performance.
English
1
0
1
1K
Samuel L Smith
Samuel L Smith@SamuelMLSmith·
The new London Training team @OpenAI has already had remarkable impact internally, alongside our phenomenal SF colleagues. I'm so excited to now see our contributions start to land in production!
OpenAI@OpenAI

GPT-5.2 Thinking evals

English
16
15
514
147.6K
Samuel L Smith ретвитнул
Sam Schoenholz
Sam Schoenholz@sschoenholz·
One of the best things in my career has been watching all the things Brain residents have gone on to do. @JeffDean, @ilyasut, Samy, Leslie, and co sure put together an amazing program. Thanks a lot!
Jeff Dean@JeffDean

I love seeing all the amazing things our Google Brain Residents and AI Residency Program cohorts have gone on to do! @hyhieu226's post spurred me to dig up a blog post and video from that era. Blog: "Google Brain Residency Program - 7 months in and looking ahead" share.google/pe5yZVCNXASL2v… Video: m.youtube.com/watch?v=KNstfq…

English
2
2
114
30.2K
Samuel L Smith
Samuel L Smith@SamuelMLSmith·
@giffmana There could literally be one researcher from each of OpenAI, GDM and Meta
English
0
0
4
871
Lucas Beyer (bl16)
Lucas Beyer (bl16)@giffmana·
It's sooo weird to me that not a single of these ~100 ex-Meta/GDM/OAI people in this new AI startup comes to my mind. Surely I must know at least a dozen or so of them? It's out of stealth now, you hid well, now show yourselves :) I mean this in all seriousness, I'm perplexed!
Sawyer Merritt@SawyerMerritt

NEWS: Jeff Bezos has created a new AI startup where he will be Co-CEO. It's called Project Prometheus and has received $6.2B in funding, some from Bezos himself. The startup is going to build AI products for engineering and manufacturing in fields like computers, aerospace and automobiles. The company already has almost 100 staff, including researchers from Meta, OpenAI and Google DeepMind.

English
23
7
415
168.5K
Samuel L Smith
Samuel L Smith@SamuelMLSmith·
@jxmnop i honestly think this is primarily just liquid vs illiquid comp
English
0
0
1
393
dr. jack morris
dr. jack morris@jxmnop·
there are dozens or perhaps a couple hundred ex-{OpenAI, xAI, Google DeepMind} researchers founding companies in the current climate there are, as far as i know, zero people leaving to found startups out of Anthropic really makes you think
English
89
49
2.2K
732.1K
Samuel L Smith
Samuel L Smith@SamuelMLSmith·
@_arohan_ Of course, in reality the batch size vs learning rate scaling had first been found in the 90s, but forgotten.
English
0
1
4
372
Samuel L Smith
Samuel L Smith@SamuelMLSmith·
@_arohan_ A reviewer for "Don't decay the learning rate, increase the batch size" asked me to compare to Hogwild. I said there was no need because people would stop using async training now, reviewer wasn't particularly happy about it but I was right! (arxiv.org/abs/1711.00489)
English
1
1
5
427
rohan anil
rohan anil@_arohan_·
Dropping a bit of Lore on this halloween that I got reminded of. Before the first TPU was taped out there was mostly async training of neural nets at Google production. The team was genuinely worried that sync training would be bad and there was a team considering figuring out how to add corruption/ asyncness so that models would converge (theory was that noise helps convergence) and was then disproven by data.
English
10
1
201
25.3K
Samuel L Smith
Samuel L Smith@SamuelMLSmith·
@prfsanjeevarora Theorists are like Physicists. They become incredibly valuable the moment they realize they are in the wrong field!
English
0
0
7
930
Sanjeev Arora
Sanjeev Arora@prfsanjeevarora·
Bravo! As a theorist at heart, I of course agree but can never know if my gut feel is biased.
Ankit Singhal@notankitsinghal

We’ve found a ton of value hiring folks with strong theory backgrounds with little to no production ML experience. One of our members of technical staff got his phd in pure math/the geometry of black holes and had no prior ML experience. Within days of hiring him we released our first model Odyssey and he had already earned himself a co-author position—that’s how invaluable he became.

English
1
2
109
20.4K
Samuel L Smith
Samuel L Smith@SamuelMLSmith·
The Training team @OpenAI is hiring researchers in London 🚀 Our twin missions are to train better LLMs, and serve them more cheaply Get in touch if you are excited to collaborate on architecture design, reliable scaling, and faster optimization
English
11
38
490
88.7K
Samuel L Smith ретвитнул
Aleksandar Botev
Aleksandar Botev@botev_mg·
If anyone is interested on working in an exciting team at the frontier of LLM research in London, please reach out to me or Sam.
Samuel L Smith@SamuelMLSmith

The Training team @OpenAI is hiring researchers in London 🚀 Our twin missions are to train better LLMs, and serve them more cheaply Get in touch if you are excited to collaborate on architecture design, reliable scaling, and faster optimization

English
3
2
15
2.6K

Открыть

@petergostev @OpenAI @flurin17 @CoachApumaYTube @JeffDean @ilyasut @giffmana @jxmnop