Pinned
My @iclr_conf 2025 Keynote on "Open-Endedness, World Models, and the Automation of Innovation" is now publicly available: iclr.cc/virtual/2025/i…
Tim Rocktäschel
5,907 posts
Tim Rocktäschel
@_rockt
Teaching AI the joy of invention at @Recursive_SI, Professor of AI @AI_UCL, PI @UCL_DARK, Fellow @ELLISforEurope. Ex @GoogleDeepMind @AIatMeta @CompSciOxford
00:00
00:06- 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.
GIF - Highly non-convex toy optimization problem. SGD vs Adam vs Evolutionary Strategies (ES). Let's gooooooo...
GIF - Harder, Better, Faster, Stronger, Real-time! We are excited to reveal Genie 3, our most capable real-time foundational world model. Fantastic cross-team effort led by @jparkerholder and @shlomifruchter. Below some interactive worlds and capabilities that were highlights for me
00:00 - Like everyone else I am extremely excited and optimistic about AI progress. But since there seems to be the misconception that it is really easy to make predictions about the future for exponential trends (let's just draw lines on a log scale), this is a reminder that while
- arxiv.org/abs/2309.16797 You're welcome.We're missing (at least one) major paradigm for LLM learning. Not sure what to call it, possibly it has a name - system prompt learning? Pretraining is for knowledge. Finetuning (SL/RL) is for habitual behavior. Both of these involve a change in parameters but a lot of humanarxiv.orgPromptbreeder: Self-Referential Self-Improvement Via Prompt EvolutionPopular prompt strategies like Chain-of-Thought Prompting can dramatically improve the reasoning abilities of Large Language Models (LLMs) in various domains. However, such hand-crafted...
- Finally, I can again include tikZ figures in my lectures, talks and papers without wasting the precious time given to me on this earth 😅
- I took Genie 3 on a paraglide over the Grand Canyon. Prompt: "A real-world paraglide through the grand canyon during sunset on a misty day."
00:00 - I am proud to announce the release of the NetHack Learning Environment (NLE)! NetHack is an extremely difficult procedurally-generated grid-world dungeon-crawl game that strikes a great balance between complexity and speed for single-agent reinforcement learning research. 1/
GIF - After seven years, I have returned to @DeepMind today 🔥 Excited about what lies ahead, and catching up with many old friends and new ones over the coming months!
- I have published an introduction to Einstein summation (einsum) notation for Deep Learning. Einsum is the Swiss Army knife for all kinds of tensor operations. It is available in @TensorFlow and with v0.4 now also in @PyTorch. rockt.github.com/2018/04/30/ein…
