AIExplained
patreon
DeepMind Prof. Tim Rocktäschel on Takeoff Speeds, GDP 2x-ing, Gemini 2, ASI timelines and Automating Science
🕑 Added 2024-12-01 15:18:53 +0000 UTCI quiz Google Deepmind Principal Scientist Prof. Tim Rocktäschel on AGI to ASI timelines, promptbreeding, GDP 2x-ing, Gemini 2 and automating science.
https://geni.us/ArtificialIntelligence
https://en.wikipedia.org/wiki/Summa_Technologiae
Laura Ruis - PROCEDURAL KNOWLEDGE IN PRETRAINING DRIVES REASONING IN LARGE LANGUAGE MODELS - https://arxiv.org/pdf/2411.12580
MLE-Bench: https://arxiv.org/pdf/2410.07095
AIDE - https://www.weco.ai/blog/technical-report
Reversal Curse: https://arxiv.org/abs/2309.12288
Reasoning or Reciting? Counter-factuals: https://arxiv.org/abs/2307.02477
Rainbow Teaming - LLM-based adversarial prompt generation - https://arxiv.org/abs/2402.16822
Sakana AI: Towards Fully Automated Open-Ended Scientific Discovery https://arxiv.org/abs/2408.06292
ChemCrow: Augmenting large language models with chemistry tools https://arxiv.org/pdf/2304.05376
BALROG: BENCHMARKING AGENTIC LLM AND VLM REASONING ON GAMES - https://arxiv.org/pdf/2411.13543
Promptbreeder - Self-referential LLM prompt evolution - https://arxiv.org/abs/2309.16797
Chain-of-Thought Prompting - Improving LLM reasoning through prompting - https://arxiv.org/abs/2201.11903
Many-shot In-Context Learning: https://arxiv.org/abs/2404.11018
NetHack Learning Environment - RL research in procedurally generated game - https://arxiv.org/abs/2006.1376
GENIE - Generative interactive environment from unlabelled videos - https://arxiv.org/abs/2402.15391
Comments
Adrian Schmidt
I agree. That bit really scared me to be honest. While I don't think the atrocities you mention should be blamed on science (I think it's more about people using science and technology for ill purposes), I certainly don't think science is a safeguard against atrocities in any way at all. It's frankly a little astonishing to me how someone might come to that conclusion… 🤔
Sean Betts
Great interview - thanks for sharing!
Arek Stryjski
Sorry I found your guest belive that science and technology always lead to good a bit ahistorical. The main witch hunts period was 15th - 17th century, so after the Middle Ages and after discovery of printing press (which unfortunately enabled it). The scientific revolution occurred well before the emergence of Stalinism and Nazism and unfortunately enabled many of the atrocities. In the past technology and science was not always beneficial. All discoveries probably had good and bad consequences. I'm not saying your arguments are wrong, but if AI starts using counterfactuals to prove some argument we will say it hallucinated. 😀 Sorry
Adin Softic
That laugh at the end 😁😆 thanks for trying to get those info On the serious note, if there is a ASI, wouldn't it also mean it could be able to model more efficient and precise biological and physical simulators? Sure some tests are still to be performed, but a lot could be anticipated
