Alexander Sasha Vezhnevets

AI research scientist at GoogleDeepMind and Vajrayana practitioner in London, UK

[Alexander Sasha Vezhnevets @Vezhnick]

Our library for Multi-Agent RL is alive and well and keeps getting better!

[Joel Z Leibo @jzl86]

Melting Pot updated! Happy to announce that our dep Lab2d now has an updated wheel for modern environments, and we've updated meltingpot accordingly.

So your agentic coding tool no longer needs to run an old python version in a sandbox when it runs meltingpot :)

[Alexander Sasha Vezhnevets @Vezhnick]

If you don't do multi-agent, multi-agent does you!

[Henry Shevlin @dioscuri]

It’s easy to talk about AI for global good, but what does benefitting everyone mean in practice? What are people owed, and by whom?

In our latest DMI essay, my brilliant colleagues @IasonGabriel and @Dr_Atoosa tackle these questions head on. Highly recommend!

[Shane Legg @ShaneLegg]

Today we have 3 new DeepMind Institute essays:

How can we control misbehaviour in agent swarms?
How should we orchestrate complex networks of AIs and people?
The case for making AGI’s benefits equitably distributed.

Get them here -> https://t.co/oVuHTDUnvD

[Atoosa Kasirzadeh (in Berkeley 1-6 Oct) @Dr_Atoosa]

Who owns the benefits of AI? In this essay I co-authored with @IasonGabriel for the DeepMind Institute, we offer four moral arguments — grounded in rights, reciprocity, fairness, and beneficence — that together make the philosophical case that the benefits of AI should reach everyone.

Read the piece here and let us know what you think: https://t.co/whTJb6oZlg

[Alexander Sasha Vezhnevets @Vezhnick]

Could agentic swarms govern themselves through Elinor Ostrom's knowledge commons framework? Our new essay from @PaglieriDavide and myself for the DeepMind Institute is out today: [https://t.co/LQkZnCVjup]

[Karthik Tadepalli @karthiktadepall]

Really nice report. Follow up: why has the fall in AI prices been so fast?

When you plot the price decline against cumulative R&D investment rather than time, you get the elasticity of price declines to R&D investment. By this margin, AI is not unusual – its price elasticity to R&D investment is squarely in the middle of Epoch's considered technologies.

So the AI price fall is historically unprecedented because we've dumped money into AI R&D at a historically unprecedented rate – and that R&D has paid off at a very average rate.

[Gary Bradski @grbradsk]

Prepare 2B scared. Robot wiggles rope, then zero shot uses it as a whip. https://t.co/3kshkjeNTO

[Alexander Sasha Vezhnevets @Vezhnick]

Do we need to work for meaning and dignity or is absence of work a liberation? Excellent work on the topic from my colleagues in the DMI!

[Stephanie Chan @scychan_brains]

Some believe that we need work for purpose and dignity, while others see freedom from work as utopic.

New paper: What does the *empirical evidence* tell us about work and wellbeing? And what does that imply for AI futures?

[Davide Paglieri @PaglieriDavide]

BALROG’s leaderboard has three new entries, courtesy of @creus_roger

Very interesting that GPT 5.6 Sol at max effort is still within error of Gemini 3 and 3.1 pro. GPT Astra 6 on max reasoning effort however reaches new heights, and a @NetHack_LE avg. progression of 13% 🏰

[Alexander Sasha Vezhnevets @Vezhnick]

Completely agree with Gillian's take on our paper. We believe another important aspect was "scientific conference" framing that agents received, which activated latent norms of scientific integrity.

[Gillian Hadfield @ghadfield]

I spoke to @amitkatwala at @techreview about Google DeepMind's new swarm experiment, where agents solving math problems started cheating and other agents blew the whistle on them.

The key point: these agents had official channels to talk to each other, and that may be what contributed to enforcement behavior we didn’t see in the Hugging Face incident. This converges with my view that alignment is institutional, not (just) dispositional. We try to train people to be good and kind, but what we really rely on is consequences if you step out of line.
https://t.co/4WZjCdCbb6

[Alexander Sasha Vezhnevets @Vezhnick]

Enjoyed MIT tech review's coverage of our latest paper
https://t.co/gpmIfYUBry

[Alex Imas @alexolegimas]

Today we are announcing the new DeepMind Institute, a forum dedicated to interdisciplinary, evidence-led debate on the societal and economic questions surrounding AGI.

As part of the launch, @JulianDJacobs and I have a new essay and working paper: “Economic Policy for AGI.”

https://t.co/cCSdKMZneH 🧵1/n

[Alexander Sasha Vezhnevets @Vezhnick]

Watch this space!

[Demis Hassabis @demishassabis]

For 20+ years @ShaneLegg and I've discussed AGI’s potential impact on the economy, science & society. With the DeepMind Institute, we're expanding interdisciplinary research on key questions for the AI era. We hope it spurs the discussions needed to get the next steps right: https://t.co/Bqf10G3Xep

[Alexander Sasha Vezhnevets @Vezhnick]

RT @alexolegimas: Thrilled to release AI in Science: Early Insights into the world. This was a project co-led with the excellent @JMateosGa…

[David Deutsch @DavidDeutschOxf]

I just fixed my dishwasher with the help of ChatGPT. A trivial task. I had been about to order a new one. So this software has increased the country's real wealth yet decreased the measured GDP. The main economic indicator is structurally incapable of registering the thing that actually makes people better off, namely the growth of knowledge

[Alexander Sasha Vezhnevets @Vezhnick]

After a long pause I'm reviving my account! Sharing our latest work with @PaglieriDavide @locross @weballergy @jzl86. Our theoretical instinct is to cast the problem of managing the agentic swarms' shared infrastructure as the knowledge commons governance.

[Davide Paglieri @PaglieriDavide]

🧵We conducted an experiment with 100 agents by giving them math problems to solve. When a small group (9%) started to cheat, what came next surprised us: 24% fought back by blowing the whistle on their peers and alerting humans.

[Alex Imas @alexolegimas]

Timely work from my @GoogleDeepMind colleagues.

[Matija Franklin @FranklinMatija]

Highly recommend checking out this new work from my colleagues. In an experiment evaluating 100 autonomous agents on math tasks, they observed that after 9% engaged in dishonest behaviour, roughly 24% resisted - actively reporting the misconduct and escalating the issue to human monitors.

[Alexander Sasha Vezhnevets @Vezhnick]

@clairlemon Neither in the UK, Canada or New Zealand. Anglosphere seems not to have this problem.