Tom Erez

Hello all! 👋 🚨 New Preprint Alert! 🚨

Code World Models for General Game-Playing. ♟️🎲 ♣️♥️♠️♦️

I am pleased to announce our new paper, which provides an extremely sample-efficient way to create an agent that can perform well in multi-agent, partially-observed, symbolic environments!

🧵 1/N

Fermentation technology as a driver of human brain expansion - Communications Biology As brain tissue is metabolically expensive, the evolutionary expansion of human brains required shifts in energy intake and metabolism. The External Fermentation Hypothesis proposes that a shift to co...

A recent theory suggests that fermentation is the original magic, because brains needed to grow (thanks to the improved nutrition) before we could safely handle fire. www.nature.com/articles/s42...

This is the most interesting contribution to the discussion about consciousness I've read in years. For me, the discussion is pretty much resolved.

Gemma explained: What’s new in Gemma 3- Google Developers Blog Google's Gemma 3 model includes vision-language support and architectural changes for resource-friendly multimodal language models.

Gemma 3 explained: Longer context, image support, and a new 1B model. → goo.gle/4lV8iaw

Other key enhancements:
🔸 Best model that fits in a single consumer GPU or TPU host
🔸 KV-cache memory reduction with 5-to-1 interleaved attention
🔸 And more!

Read the blog for the full details on Gemma 3.

Today we are releasing the dataset of table tennis ball trajectories used to train the Google DeepMind robot that can play amateur table tennis with humans (sites.google.com/corp/view/co...). This work was accepted for presentation at #ICRA2025 and we hope to see you there!

YouTube video by Google for Developers Inside Gemma 3: Modifying the output through activation hacking

Gemma 3 are just amazing models!

but what if you want to manipulate it's internal activations to understand how it does its text generation?

Sascha Rothe is here to teach you how!

Great insights for anyone curious about the inner workings of LLMs!

www.youtube.com/watch?v=JTUs...

“Wanting to be Understood” - Could a deep human need to be understood be the crucial evolutionary 'gadget' bootstrapping cooperation, culture, and language? We explore this idea using AI simulations in our new paper: arxiv.org/abs/2504.06611 🧠
#Evolution #Cognition #AI #GoogleDeepMind

New paper from our team @GoogleDeepMind!

🚨 We've put LLMs to the test as writing co-pilots – how good are they really at helping us write? LLMs are increasingly used for open-ended tasks like writing assistance, but how do we assess their effectiveness? 🤔

arxiv.org/pdf/2503.19711

🚨 I’m hosting a Student Researcher @GoogleDeepMind!

Join us on the Autonomous Assistants team (led by
@egrefen.bsky.social) to explore multi-agent communication—how agents learn to interact, coordinate, and solve tasks together.

DM me for details!

🥁Introducing Gemini 2.5, our most intelligent model with impressive capabilities in advanced reasoning and coding.

Now integrating thinking capabilities, 2.5 Pro Experimental is our most performant Gemini model yet. It’s #1 on the LM Arena leaderboard. 🥇

Don't lie to your friends: Learning what you know from collaborative self-play To be helpful assistants, AI agents must be aware of their own capabilities and limitations. This includes knowing when to answer from parametric knowledge versus using tools, when to trust tool outpu...

We all want LLMs to collaborate with humans to help them achieve their goals. But LLMs are not trained to collaborate, they are trained to imitate. Can we teach LM agents to help humans by first making them help each other?

arxiv.org/abs/2503.14481

Looking for a small or medium sized VLM? PaliGemma 2 spans more than 150x of compute!

Not sure yet if you want to invest the time 🪄finetuning🪄 on your data? Give it a try with our ready-to-use "mix" checkpoints:

🤗 huggingface.co/blog/paligem...
🎤 developers.googleblog.com/en/introduci...

How do we ensure humans can still effectively oversee increasingly powerful AI systems? In our blog, we argue that achieving Human-AI complementarity is an underexplored yet vital piece of this puzzle! And, it’s hard, but we achieved it.

🧵(1/10)

Making LLMs run efficiently can feel scary, but scaling isn’t magic, it’s math! We wanted to demystify the “systems view” of LLMs and wrote a little textbook called “How To Scale Your Model” which we’re releasing today. 1/n

MuJoCo Playground An open-source framework for GPU-accelerated robot learning and sim-to-real transfer

Introducing playground.mujoco.org
Combining MuJoCo’s rich and thriving ecosystem, massively parallel GPU-accelerated simulation, and real-world results across a diverse range of robot platforms: quadrupeds, humanoids, dexterous hands, and arms.
Get started today: pip install playground

Apptronik Partners with Google DeepMind.
apptronik.com/news-collect...

Would it make sense to add this account (for MuJoCo announcements etc.)?

I'd be happy to be included (MuJoCo news etc.)

New paper! We show that by using keypoint-based image representation, robot policies become robust to different object types and background changes.

We call this method Prescriptive Point Priors for robot Policies or P3-PO in short. Full project is here: point-priors.github.io

HOT 🔥 fastest, most precise, and most capable hand control setup ever...

Less than $450 and fully open-source 🤯
by @huggingface, @therobotstudio, @NepYope

This tendon-driven technology will disrupt robotics! Retweet to accelerate its democratization 🚀

A thread 🧵

Check out Motivo, a behavioral foundation model for humanoid control by FAIR.

It's a one-of-its-kind unsupervised RL project, and it comes with a demo that is SO fun to play with!

metamotivo.metademolab.com

(for the record, they use compile and cudagraphs -> github.com/facebookrese...)

🚀🚀PaliGemma 2 is our updated and improved PaliGemma release using the Gemma 2 models and providing new pre-trained checkpoints for the full cross product of {224px,448px,896px} resolutions and {3B,10B,28B} model sizes.

1/7

Probabilistic weather forecasting with machine learning - Nature GenCast, a probabilistic weather model using artificial intelligence for weather forecasting, has greater skill and speed than the top operational medium-range weather forecast in the world and provid...

The next generation of probabilistic machine learning for weather called GenCast is published in @natureportfolio.bsky.social today 🥳. Amazing to see the collective progress in ML for weather as a field over the last 5 years. 🏖️ www.nature.com/articles/s41...

AI for science could be more impactful than chatbots. It is already helping win Nobel prizes and accelerating drug development and materials discovery.
Today we published an essay about it: why it matters, how it’s happening and its implications. Here is a summary from an econ / social sci lens.

MuJoCo meetup at CoRL 2024 · google-deepmind mujoco · Discussion #2203 Hello MuJoCo community! If you are in CoRL 2024 (Munich, Germany), we will be hosting an informal meetup during the 1-2pm lunch break on Friday, November 8th, at the Congress Center's Mars room Swi...

If you're coming to Munich for CoRL this week, come say hi!
github.com/google-deepm...

I'm in a starter pack!
go.bsky.app/GZ4hZzu

I want to take care of everybody and I'm furious at the people who don't want to take care of everybody but I need to not step off the everybody plate just because they suck

"A plausible explanation of this is that persons who have achieved or enjoy high social status are less willing to entertain the possibility that a disaster could occur which would spoil everything."