Samuel Lavoie

AI scientist @google. Amortizing compute and learning representations.

[Samuel Lavoie @lavoiems]

🧵 Everyone is chasing new diffusion models—but what about the representations they model from?
We introduce Discrete Latent Codes (DLCs):
- Discrete representation for diffusion models
- Uncond. gen. SOTA FID (1.59 on ImageNet)
- Compositional generation
- Integrates with LLM
🧱

[Daniel Sinclair @_DanielSinclair]

There's a great clip here.

"It was the first car ever designed by you, right?"

"🤬🤬 yea, it...."

[Samuel Lavoie @lavoiems]

@mavenlin Exciting news Min!

[Michael Noukhovitch @mnoukhov]

I rambled somewhat-coherently on a podcast about discrete representations for diffusion models
(https://t.co/wzHBfX2eUK) and recent MARL work solving AI social dilemmas
(https://t.co/hgiqkUFHp9)

Thank you @joycech3n for the great convo and laughing at my one mediocre joke

[Samuel Lavoie @lavoiems]

@benno_krojer Great post! It’s great to read how your perception of ai science has evolved!

[Michael Noukhovitch @mnoukhov]

Olmo 3.1: even more RL = even more RL-Zero!

@saurabh_shah2 and I tweaked some hyperparams and prompts, @hamishivi and @finbarrtimbers improved the code and boom!

New Olmo 3.1 RL-Zero 👾 An updated, solid baseline for your RL and reasoning research

[Michael Galkin @michael_galkin]

I'll be giving a talk 20 min after Jure at the graph learning workshop, the most GFM-packed event at neurips so far 👏

[Michael Noukhovitch @mnoukhov]

We've printed stickers for our #NeurIPS2025 poster. Come check it out tomorrow (Friday) at 11am to get one, poster #3505

[Muqeeth @Muqeeth10]

New preprint! Learning Robust Social Strategies with Large Language Models. We apply multi-agent RL finetuning to train LLMs that achieve cooperative and non-exploitable behavior in social dilemmas for the first time.

📄 https://t.co/lMKxJ4XoBx
🧵 ⬇️
(1/8)

[Anirudh Buvanesh @AnirudhBuvanesh]

Zero rewards after tons of RL training? 😞 Before using dense rewards or incentivizing exploration, try changing the data. Adding easier instances of the task can unlock RL training. 🔓📈To know more checkout our blog post here: https://t.co/BPErVcLmP8. Keep reading 🧵(1/n)

[Samuel Lavoie @lavoiems]

I'll be at #neurips next week to present this work with @mnoukhov and @AaronCourville !

[Samuel Lavoie @lavoiems]

🧵 Everyone is chasing new diffusion models—but what about the representations they model from?
We introduce Discrete Latent Codes (DLCs):
- Discrete representation for diffusion models
- Uncond. gen. SOTA FID (1.59 on ImageNet)
- Compositional generation
- Integrates with LLM
🧱

[Miguel Angel Bautista @itsbautistam]

I'm looking for reviewers for TMLR around geometry-aware molecular generation that focuses on order-agnostic autoregressive modeling, if you have expertise in geometric deep learning or generative AI for drug discovery, please let me know! 🧬

[Michael Noukhovitch @mnoukhov]

Olmo 3 is a big release with so much cool stuff! RL Zero was super fun to work on and please check out all the cool RL infra details 😎

[Samuel Lavoie @lavoiems]

@luisenp Sorry to hear that Luis. Best of luck!

[Samuel Lavoie @lavoiems]

@francoisfleuret I think you will like some of our related works: https://t.co/dVCpOaZbHe and https://t.co/r7ZRYVkxNR

[Samuel Lavoie @lavoiems]

@oh_that_hat @hugo_larochelle Congrats Hattie!!

[Samuel Lavoie @lavoiems]

@WalterMayor_T @johanobandoc @AaronCourville @pcastr Great work Walter!

[Walter Mayor-Toro @WalterMayor_T]

Proud to have contributed, as an independent researcher, to the great work achieved by our research team.
Many thanks to @johanobandoc , @lavoiems , Scott Fujimoto, @AaronCourville and @pcastr for making it happen.
See Pablo’s thread for more details.

[Johan Obando-Ceron 👍🏽 @johanobandoc]

1/3 🥳Excited to share our new paper ‘Simplicial Embeddings Improve Sample Efficiency in Actor–Critic Agents’!

Project your features onto a product of simplices → sparse, stable reps, stronger grads, faster learning.

🧵For more details, check out Pablo’s thread 👇

[Samuel Lavoie @lavoiems]

It was awesome collaborating on that project. We show why SEM improves RL training of RL agents 🚀.

[Pablo Samuel Castro @pcastr]

🔊Simplicial Embeddings (SEMs) Improve Sample Efficiency in Actor-Critic Agents🔊

In our recent preprint we demonstrate that the use of well-structured representations (SEMs) can dramatically improve sample efficiency in RL agents.

1/X

[Pablo Samuel Castro @pcastr]

🔊Simplicial Embeddings (SEMs) Improve Sample Efficiency in Actor-Critic Agents🔊

In our recent preprint we demonstrate that the use of well-structured representations (SEMs) can dramatically improve sample efficiency in RL agents.

1/X