Samuel L Smith

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

[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

[Nikolay Savinov @SavinovNikolay]

Excited to share that I’ve joined OpenAI in London to work on pretraining!

I’ve spent the last few years on pretraining and long-context, and I’ll be helping grow the London team. We’re hiring exceptional researchers, please reach out: [email protected]

[Aidan Clark @_aidan_clark_]

Extremely excited to work with Nikolay!

The OpenAI London team has led huge wins on every model release we've had in the last year, and is all-around awesome :) Nikolay is an incredible addition, and if you're interested in joining, please reach out to him or I!

[roon @tszzl]

one take that always ages really badly is that the models are intelligent enough and nobody can really notice the improvements anymore

[Samuel L Smith @SamuelMLSmith]

@nanomader @pfau @LucaAmb I was talking as a researcher. I just think models still have a lot of room to improve and we have plenty more work to do!

[Stanislav Fort @stanislavfort]

Anthropic chose FreeBSD to showcase their Mythos zero-days. In the latest release, 8 CVEs were announced:
3 found by Anthropic, 3 discovered by AISLE's AI (!)

AISLE is matching Mythos 3-for-3 on zero-days on the very codebase of their choosing at a fraction of the cost

[AI Security Institute @AISecurityInst]

OpenAI’s GPT-5.5 is the second model to complete one of our multi-step cyber-attack simulations end-to-end 🧵

[Samuel L Smith @SamuelMLSmith]

@petergostev @OpenAI the research team in London is outperforming expectations and growing fast!

[Samuel L Smith @SamuelMLSmith]

A big win born in @OpenAI London! 🇬🇧

[OpenAI Developers @OpenAIDevs]

GPT-5.2 and GPT-5.2-Codex are now 40% faster.

We have optimized our inference stack for all API customers.

Same model. Same weights. Lower latency.

[Samuel L Smith @SamuelMLSmith]

@flurin17 @OpenAI no!

[Samuel L Smith @SamuelMLSmith]

@CoachApumaYTube @OpenAI Entirely inference optimizations

[Greg Brockman @gdb]

lots of exciting work happening in our london office!

[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

[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!

[Samuel L Smith @SamuelMLSmith]

@giffmana There could literally be one researcher from each of OpenAI, GDM and Meta

[Samuel L Smith @SamuelMLSmith]

@jxmnop i honestly think this is primarily just liquid vs illiquid comp

[Samuel L Smith @SamuelMLSmith]

Pretty sure we already have 😂

[Ellis Brown @_ellisbrown]

🌶️ hot take 🌶️
> we should normalize training on the test set

yes, you read that right.
no, I'm not joking.
and, yes... I have taken ML 101

👉 here's why this is crucial for future multimodal LLM research [1/n] 🧵

[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.

[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! (https://t.co/brxMEypOUY)