🔥Telegram İfşa

I think people are overindexing on the OpenAI o3 ARC-AGI results. There’s a long history in AI of people holding up a benchmark as requiring superintelligence, the benchmark being beaten, and people being underwhelmed with the model that beat it.

Ivana Balazevic Felix Hill I know several others in tech who had psychosis triggered by drugs seen as “safe”. I’ve known multiple people who died in the aftermath. Tech drug culture, especially in Silicon Valley, is a problem nobody sees as a problem.

I’ve sadly witnessed Felix Hill’s life unravel over the past 2 years. So incredibly heartbreaking. Sharing his story to raise awareness of the dangers of ketamine and to reduce the stigma around mental health issues. You have only one brain folks, please take care of it.

The RE application is now open: boards.greenhouse.io/deepmind/jobs/… And here is the link to the RS position: boards.greenhouse.io/deepmind/jobs/…

this is the most california thing ive ever seen

Lots of vague AI hype on social media these days. There are good reasons to be optimistic about further progress, but plenty of unsolved research problems remain.

After an amazing 6 years at Google DeepMind, I'm thrilled to announce that I'll be starting a new project at the intersection of multimodal foundation modeling, data curation, and human behavior. If this is of interest to you please reach out!

Yeah they gotta do it themselves now since trump fired like 90% of the national park service

One thing I notice in SF is the blurred lines between work and play time e.g. people performative coding at bars / parties signaling “hustle”. This is extremely peasant-coded and not conducive to great work imo. History’s elites and greats kept strict borders and took leisure

(3/4) Context-Aware Multimodal Pretraining By the amazing Karsten Roth with Zeynep Akata, Dima Damen @ICVSS2025, Ivana Balazevic, Olivier Hénaff 📖 [Paper]: arxiv.org/abs/2411.15099

cooking up something tasty for tomorrow...

As I was saying: it's happening

Got access to Google diffusion. HOLY SH!T 909 tokens/s ?????? I made a calendar in 3s? 3 fcuking seconds?

guys this is real time

In 2022, I worked on text diffusion for a bit and wrote a blog post. Since then, people have regularly asked me about scaling diffusion LLMs. All the while, I was on the first row watching Brendan assemble a cracked team and make it a reality. Now I can stop being coy about it😁

To add to the buzz of this week, a bit more of a personal update: I’m moving to San Francisco in a few days! Grateful for the 3.5 amazing years in London and all the wonderful friends and colleagues I’ve met here. ♥️ I’ll be staying at Google DeepMind and happy to chat about

Writing a STRONGLY WORDED LETTER with Gemini Diffusion and its Instant Edit tab

In Nashville for my last PhD conference 🥲. Come join today 10:30-12:30 in Hall D (#391) to talk insights, tips and tricks to modify pretraining for representation reuse - scalably. 🚀Joint work w/ Zeynep Akata, Dima Damen @CVPR 2025, Ivana Balazevic & Olivier Hénaff while at Google DeepMind.