开启时会模糊预览图，关闭后正常显示
![LeCun’s new company on physical AI with world models [9] looks a lot like our 2014 company on physical AI with world models [1] 😀 See also [2-8] - all references in the reply!](/_next/image?url=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FG8tMtBCXEAAcnDn.jpg&w=3840&q=75 "LeCun’s new company on physical AI with world models [9] looks a lot like our 2014 company on physical AI with world models [1] 😀 See also [2-8] - all references in the reply!")
LeCun’s new company on physical AI with world models [9] looks a lot like our 2014 company on physical AI with world models [1] 😀 See also [2-8] - all references in the reply!
REFERENCES [1] NNAISENSE, the AGI company for AI in the physical world, founded in 2014, based on neural network world models (NWMs). J. Schmidhuber (JS) was its President and Chief Scientist - see his NWM papers 1990-2015, e.g., [4-5], and the 2020 NNAISENSE web page in the Internet Archive https://t.co/j6xsLXHdPs (Lately, however, NNAISENSE has become less AGI-focused and more specialised, with a focus on asset management.) [2] JS, AI Blog (2022). LeCun's 2022 paper on autonomous machine intelligence rehashes but does not cite essential work of 1990-2015. https://t.co/byn3K3aSxK Years ago, JS published most of what LeCun calls his "main original contributions:" neural nets that learn multiple time scales and levels of abstraction, generate subgoals, use intrinsic motivation to improve world models, and plan (1990); controllers that learn informative predictable representations (1997), etc. This was also discussed on Hacker News, reddit, and in the media. LeCun also listed the "5 best ideas 2012-2022" without mentioning that most of them are from JS's lab, and older. Popular tweets on this: https://t.co/kn7KhFHLvw https://t.co/FxALILsNRu https://t.co/caTuctmztu https://t.co/Rpip8HBzPA [3] How 3 Turing awardees republished key methods and ideas whose creators they failed to credit. Technical Report IDSIA-23-23, Swiss AI Lab IDSIA, 2023. https://t.co/Nz0fjc6kyx Best start with Section 3. See also [8]. Popular tweet on this: https://t.co/0fJVklXyOr [4] JS (1990). Making the world differentiable: on using fully recurrent self-supervised neural networks for dynamic reinforcement learning and planning in non-stationary environments. TR FKI-126-90, TUM. This report used the terminology "world model” for a recurrent neural network that learns to predict the environment and the consequences of the actions of a separate controller neural net. It also introduced "artificial curiosity" and "intrinsic motivation" through generative adversarial networks. Led to lots of follow-up publications. [4b] JS (2002). Exploring the Predictable. In Ghosh, S. Tsutsui, eds., Advances in Evolutionary Computing, p. 579-612, Springer, 2002. Don't predict pixels - find predictable internal representations / abstractions of complex spatio-temporal events! [5] JS (2015). On Learning to Think: Algorithmic Information Theory for Novel Combinations of RL Controllers and Recurrent Neural World Models. ArXiv 1210.0118. Introducing a reinforcement learning (RL) prompt engineer and adaptive chain of thought: an RL neural net learns to query its "world model" net for abstract reasoning & decision making. Going beyond the 1990 neural world model [4] for millisecond-by-millisecond planning. See tweet for 10-year anniversary: https://t.co/3FYt4x2PMM [6] JS (2018). One Big Net For Everything. arXiv 1802.08864. Collapsing the reinforcement learner and the world model of [5] (e.g., a foundation model) into a single network, using JS's neural network distillation procedure of 1991. See DeepSeek tweet: https://t.co/HIVU8BWAaS [7] David Ha & JS. World Models. NeurIPS 2018. https://t.co/RrUNYSIz6n [8] Who invented convolutional neural networks? Technical Note IDSIA-17-25, IDSIA, 2025. https://t.co/HdCanIa4MN Popular tweets on this: https://t.co/6eDUT8qcNE https://t.co/chfcmk253b https://t.co/h27y6Ni2CA https://t.co/Rpip8HBzPA LinkedIn https://t.co/vzKQPhAGAy [9] Sifted dot eu (18 Dec 2024). Yann LeCun raising €500m at €3bn valuation for new AI startup. The outgoing Meta exec announced last month he was launching a new project to build “world models.” https://t.co/c21tW6sy3b Quote: "The new company will focus on “world models”, systems that can understand the physical world instead of merely generating text like today’s large-language models (LLMs)." See [1].
Man I love Czechs talking about how Russia is "cucked" and then poasting this about their security guarantor in the face of Russian threat. They are so cute. For them, nothing was grander than this "Western Civilization" story, and so they're confused now. What's happening?!
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
RT @Sriraam_UTD: Fantastic insights! Consistency of traces being more important than correctness is quite interesting!
AI researcher & teacher @SCAI_ASU. Former President of @RealAAAI; Chair of @AAAS Sec T. Here to tweach #AI. YouTube Ch: https://t.co/4beUPOmf6y Bsky: rao2z
RT @VenkRamaswamy: @rao2z thinks LRMs are better thought of as Large Mumbling Models
AI researcher & teacher @SCAI_ASU. Former President of @RealAAAI; Chair of @AAAS Sec T. Here to tweach #AI. YouTube Ch: https://t.co/4beUPOmf6y Bsky: rao2z
I'm trying something new with my blog, making it interactive first article is about something I care deeply about: logs logging sucks so much https://t.co/mdFhxEvpjB
building workers observability @cloudflaredev, prev founder @baselimehq (acquired by cloudflare), prev aerodynamicist
I'm trying something new with my blog, making it interactive first article is about something I care deeply about: logs logging sucks so much https://t.co/mdFhxEvpjB
building workers observability @cloudflaredev, prev founder @baselimehq (acquired by cloudflare), prev aerodynamicist