Lei M. Zhang

Cool paper using cultural learning to better align culture values in LLMs. Cultural learning ftw! 💪👨‍🏫

Researchers Stephen Jordan and Noah Shutty from Google Quantum AI have devised a new quantum approach to solving problems by treating them like error-free messages. So far, the algorithm has bested all known classical techniques. www.quantamagazine.org/quantum-spee...

Best practices... no wait...
Good enough practices... err... how about
Questionable practices in machine learning https://arxiv.org/abs/2407.12220

A brilliant colleague and wonderful soul Felix Hill recently passed away. This was a shock and in an effort to sort some things out, I wrote them down. Maybe this will help someone else, but at the very least it helped me. Rest in peace, Felix, you will be missed. www.janexwang.com/blog/2025/1/...

Excited to be able to share what I've been up to recently!

We released a new experimental model, Gemini 2.0 Flash Thinking, which shows the thoughts/reasoning it uses to come up with its answers.

Try it out here! aistudio.google.com/prompts/new_...

Try out our fresh-from-the-oven thinking model: aistudio.google.com/prompts/new_...

Very cool what these new classes of models are capable of!

Acheulean technology and emergent sociality: what material engagement means for the evolution of human-environment systems - Phenomenology and the Cognitive Sciences The Acheulean techno-complex represents a significant chapter in hominin cognitive evolution. Two important developments include enchained technological behaviours that were practiced with broad consi...

Acheulean technology and emergent sociality: what material engagement means for the evolution of human-environment systems link.springer.com/article/10.1...

It’s been a crazy 2 years seeing so many amazingly talented researchers bring GENerative Interactive Environments alive in Genie 1 and 2.

the future is agents in generative environments

Introducing 🧞Genie 2 🧞 - our most capable large-scale foundation world model, which can generate a diverse array of consistent worlds, playable for up to a minute. We believe Genie 2 could unlock the next wave of capabilities for embodied agents 🧠.

thinking of calling this "The Illusion Illusion"

(more examples below)

Boundless Socratic Learning with Language Games An agent trained within a closed system can master any desired capability, as long as the following three conditions hold: (a) it receives sufficiently informative and aligned feedback, (b) its covera...

Are there limits to what you can learn in a closed system? Do we need human feedback in training? Is scale all we need? Should we play language games? What even is "recursive self-improvement"?

Thoughts about this and more here:
arxiv.org/abs/2411.16905

we are recruiting interns for a few projects with @pyoudeyer
in bordeaux
> studying llm-mediated cultural evolution with @nisioti_eleni
@Jeremy__Perez

> balancing exploration and exploitation with autotelic rl with @ClementRomac

details and links in 🧵
please share!

Applications are open for student researcher positions at Google DeepMind for 2025! Due date Dec 13

www.google.com/about/career...

During most of the vast history of humanity, technologies have evolved gradually, not through leaps of genius, but slow cultural adaption.

Before the scientific method, easy communication & printed books, innovation happened at the level of societies, not people. www2.psych.ubc.ca/~henrich/pdf...

The lucky author holds a single, roasted cacao bean with 1/2 the shell removed so we can see the pure chocolate inside. Behind his hand is a roasting tray filled with beans. The smell. THE SMELL.

This photo shows an aluminum roasting tray filled with a layer of cacao beans. When roasting, the aroma of the fermentation suddenly gives way to the overwhelming joy of fresh brownies. Yes, there is a spirit watching over us all.

Reader, I do not know the demons that haunt your dreams, the frustrations and insults of life that have brought you to your knees, the sadness you battle in your heart.

But know this: there is pure goodness in this world. It exists. Sometime we see a glimpse of it.

Today, I got such a peek.

Just a heads up to everyone: @deep-mind.bsky.social is unfortunately a fake account and has been reported. Please do not follow it nor repost anything from it.

Great list of researchers working on making NLP models more culturally aware 🌏!!

Google DeepMind Starter Pack Join the conversation

For no particular reason, I really like these starter packs:

Google DeepMind: go.bsky.app/GZ4hZzu
Theoretical CS: go.bsky.app/6A6GRSi
ML Theory: go.bsky.app/21nFz12
Differential Privacy: go.bsky.app/A7ABG83

Human-Timescale Adaptation in an Open-Ended Task Space Foundation models have shown impressive adaptation and scalability in supervised and self-supervised learning problems, but so far these successes have not fully translated to reinforcement learning (...

We tried something similar in adaptive agents arxiv.org/abs/2301.07608 (multiturn adaptation + RL). This was with transformers but not LLMs. Someone should try this with LLMs!

now, if we think of p(output | prompt, a few examples) as a predictive distribution p(y|x, D) ... it looks very much like learning to me :)

see e.g. my slide deck on drive.google.com/file/d/1B-Ka...

Learning high-accuracy error decoding for quantum processors - Nature A recurrent, transformer-based neural network, called AlphaQubit, learns high-accuracy error decoding to suppress the errors that occur in quantum systems, opening the prospect of using neural-network...

Beyond excited to introduce AlphaQubit, now published in
Nature!

AlphaQubit is a neural network for quantum error correction and achieves state-of-the-art accuracy on simulated and real-world data.

Quantum computing: physics–AI collaboration quashes quantum errors Machine-learning strategy for quantum error correction.

I'm super excited to share that our work on AlphaQubit, a high accuracy neural network quantum error correction decoder, is now published in Nature! www.nature.com/articles/d41...

Learning high-accuracy error decoding for quantum processors - Nature A recurrent, transformer-based neural network, called AlphaQubit, learns high-accuracy error decoding to suppress the errors that occur in quantum systems, opening the prospect of using neural-network...

Really excited to share our (Google DeepMind and Google QuantumAI) new publication in @Nature, where we use neural networks to identify errors in quantum computers with state-of-the-art accuracy.
Paper: www.nature.com/articles/s41...
Blog: blog.google/technology/g...

Now that @jeffclune.bsky.social and @joelbot3000.bsky.social are here, time for an Open-Endedness starter pack.

go.bsky.app/MdVxrtD

Google DeepMind is hiring Student Researchers in EMEA 👇

Student Researcher positions in EMEA now accepting applications!

Please repost.

www.google.com/about/career...

Research Engineer, Generative AI London, UK

Our team is looking for an ML hacker extraordinaire. make the TPUs hum with the sound of AGI. Job description: make intelligence = f(compute) unbounded : boards.greenhouse.io/deepmind/job...