John Reid
@__Reidy__
Followers
356
Following
1K
Media
15
Statuses
520
Generating AI one bit at a time @DeepMind, formerly @Blue_Prism and @mrc_bsu. All ill-considered thoughts my own neurons' work and not my employers'.
London, UK
Joined May 2011
Was great working with such a talented team on embodied AGI. Check out the blog post to see exactly what it can do.
SIMA 2 is our most capable AI agent for virtual 3D worlds. 👾🌐 Powered by Gemini, it goes beyond following basic instructions to think, understand, and take actions in interactive environments – meaning you can talk to it through text, voice, or even images. Here’s how 🧵
0
0
12
How well do foundation models explore dynamic interactive environments? We look at this and more in: https://t.co/RVy5GzHACX Thanks to my co-authors including @dannypsawyer, @rosemary_ke, @martinengelcke, @AlexLerchner, @DaniloJRezende, @countzerozzz, @mc_mozer and @janexwang
Happy to announce that our work has been accepted to workshops on Multi-turn Interactions and Embodied World Models at #NeurIPS2025! Frontier foundation models are incredible, but how well can they explore in interactive environments? Paper👇 https://t.co/8Q9j1VMTYv 🧵1/13
0
1
7
1/ Today we are releasing Gemini 2.5 Pro Experimental, our newest Gemini model with integrated “thinking” and significant performance gains. Very proud of the whole team! 🧵
9
21
458
I’m really sad that my dear friend @FelixHill84 is no longer with us. He had many friends and colleagues all over the world - to try to ensure we reach them, his family have asked to share this webpage for the celebration of his life: https://t.co/1QoyHmAD3p
111
86
737
Santa Arrived! The PDF (of a colorful Xmas version) of the "Introduction to Universal AI" book is now freely available online at https://t.co/r9COEBnf4S Wishing you all joyful reading, Merry Xmas & a :-) New Year.
21
198
1K
A chess set inspired by art deco 🎨, plants 🌱, or even the universe 🌌. Our latest AI experiment GenChess in @LabsDotGoogle lets you create playable art pieces with the power of Imagen 3. Try it now → https://t.co/9th9zNoVBR
14
44
246
Beautiful result. A grandmaster-level chess LLM that just generates the next move - no search! All that was required was training a 270M parameter model on 10M chess games annotated by Stockfish.
Google Deepmind trained a grandmaster-level transformer chess player that achieves 2895 ELO, even on chess puzzles it has never seen before, with zero planning, by only predicting the next best move, if a guy told you "llms don't work on unseen data", just walk away
3
7
27
Since GPT3, we have known that large Transformer language models are effective few-shot learners. But did you know that the nature of language itself makes language particularly rich pre-training data for inducing few-shot learning? 1/n
9
35
339
This one will be good to follow. Will be exciting to see whether variational inference will become the standard for posterior inference.
0
0
4
A very brave and candid account. Thanks @FelixHill84
Do you work in AI? Do you find things uniquely stressful right now, like never before? Haver you ever suffered from a mental illness? Read my personal experience of those challenges here: https://t.co/OFZmSizl6f
0
0
1
Terry Tao on o1 (very thoughtful & interesting, unsurprisingly):
mathstodon.xyz
I have played a little bit with OpenAI's new iteration of #GPT, GPT-o1, which performs an initial reasoning step before running the LLM. It is certainly a more capable tool than previous iterations,...
3
37
250
Boosting LLM Performance with Dynamic Skill Selection! 1/ 🚀 What if LLMs could get better at solving math problems by understanding the skills they need? We explored this idea by having LLMs identify and label the skills required for each problem.
arxiv.org
Metacognitive knowledge refers to humans' intuitive knowledge of their own thinking and reasoning processes. Today's best LLMs clearly possess some reasoning processes. The paper gives evidence...
Metacognition (“Thinking about thinking”) in LLMs 🧵/1 Humans can express sophisticated thoughts about their problem-solving processes. Can LLMs do it? Our paper shows they can! https://t.co/NnejLyVs7L And this “metacognition” can help improve LLM problem-solving.
2
5
15
How do astrological predictions hold up in a well designed experiment?
Does astrology work? We tested the ability of 152 astrologers to see if they could demonstrate genuine astrological skill. Here is how the study was designed and what we found (including a result that really surprised me): 🧵
0
0
1
We’re extremely excited to announce the NeurIPS Workshop on Bayesian Decision-making and Uncertainty: from probabilistic and spatiotemporal modeling to sequential experiment design! This will take place at NeurIPS 2024, in Vancouver, BC, Canada, either on December 14th or 15th.
1
67
385
We uploaded a v2 of our book draft "The Elements of Differentiable Programming" with many improvements (~70 pages of new content) and a new chapter on differentiable data structures (lists and dictionaries).
arxiv.org
Artificial intelligence has recently experienced remarkable advances, fueled by large models, vast datasets, accelerated hardware, and, last but not least, the transformative power of...
10
123
665
Am surprised by unexpected support! Thanks @GaryMarcus
Stop the press! Marcus has something positive to say about some news in AI! Intriguing new task tying language to a range of video games, from @GoogleDeepMind. Getting to 100% (which I assume is trivial for many teenagers) would be impressive, especially if there was
1
0
3
What I've been working on recently with @zhengdongwang, @Anya77335370, @jeffclune, @_timharley amongst many others. It is all about the goat. 🐐🐐🐐
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
0
0
10
I am really excited to reveal what @GoogleDeepMind's Open Endedness Team has been up to 🚀. We introduce Genie 🧞, a foundation world model trained exclusively from Internet videos that can generate an endless variety of action-controllable 2D worlds given image prompts.
133
551
2K
Very happy to be part of this project. Kudos to @anianruoss and @gregdeletang for doing the lion's share of the work. Although not as strong as the best engines, the agent plays powerfully with a really nice style, I hope it is just the start of another breed of chess engine.
@GaryMarcus TPUs go brrrr https://t.co/RlkNoXHcXr
0
0
5