Tim Harley

Engineering & Deep Learning Research @deepmindai

[Tim Harley @_timharley]

@Heraklines1 @sebkrier Deus ex machina, indeed.

[Tim Harley @_timharley]

Super cool World-Model results from Genie team! 🧞‍♂️

Genie 2 generates pixels of an interactive 3D-world from a prompt image.

Amazing progress in only 9 months since the release of Genie-1! 🤯

A cool cameo from our SIMA agent, which can act in this world model out-of-the-box!

[Jack Parker-Holder @jparkerholder]

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 🧠.

[Tim Harley @_timharley]

@_aidan_clark_ Hard disagree.

This is like saying noone should learn to read or write English because AI will write all text in the future.

Being able to modify AI-generated code will remain a useful skill.

Basic programming literacy is becoming more relevant and accessible, not less!

[Tim Harley @_timharley]

@NandoDF @scott_e_reed Indeed! Gato was very inspiring to us in SIMA, I remember when it started being blown away by the audacity & boldness to try to fit so many domains/modalities with one model.

Video models are a core part of SIMA's perception, thanks to the Phenaki & SPARC teams for their models!

[Tim Harley @_timharley]

@ShumingHu @yaroslavvb Implicitly, asynchrony adds momentum
https://t.co/9tO5l4pYL6

[Tim Harley @_timharley]

@j_mcgraph @yaroslavvb I did a bunch of hogwild training of various neural nets & deep RL agents in 2014-2016 (on a big CPU cluster).
Hogwild: poor utilisation of GPU compared to SGD, so slower overall; also hogwild gave worse model results and more variance across runs - making science harder.

[Tim Harley @_timharley]

Massive thanks to the amazing SIMA team, our expert data team, and the game developers & studios we have partnered with for these games:

@SatisfactoryAF @Valheimgame @GoatSimulator @teardowngame @NoMansSkyGame @WobblyLifeDev @SpaceEngineersG @HydroneerGame @StrangeLoopGame

[Tim Harley @_timharley]

So proud of what we and the team have achieved with SIMA!

In this early work, our AI agent plays a number of amazing 3D open world games - but it doesn't play to win, you can tell it what you want it to do...🧵

[Google DeepMind @GoogleDeepMind]

Introducing SIMA: the first generalist AI agent to follow natural-language instructions in a broad range of 3D virtual environments and video games. 🕹️

It can complete tasks similar to a human, and outperforms an agent trained in just one setting. 🧵 https://t.co/qz3IxzUpto

[Tim Harley @_timharley]

SIMA is early work towards a General AI Agent, but it is not yet at human-level. We're excited to see where we can take it next - more games, better performance & longer, more complex tasks.

We have work to do, but it's an exciting time to be working on Agents!

[Andrew Lampinen @AndrewLampinen]

Excited to share one of the main things I’ve been working on: scaling towards grounded language agents that can follow natural language instructions across many video game environments—using the same pixels to keyboard/mouse controls as humans do. 1/8

[Seva @SevaUT]

I continue to think the bigger threat of deepfakes is not in convincing people that fake things are real but in offering plausible suspicion that real things are fake

[Tim Harley @_timharley]

@amoufarek Haha I was just messing!

[Tim Harley @_timharley]

@bgavran3 Progressive Neural Networks might be relevant
https://t.co/5pmiCG27zD

[Tim Harley @_timharley]

@will_owen9 It's the first half of the RWC final, isn't it?

[Dileep George @dileeplearning]

this is why I dislike the term stochastic parrots.....it really underestimates parrots. 😇

[Tim Harley @_timharley]

This isn't the first time that 52% of UK voters have opted for a symbolic gesture that makes no sense in practice...