NeuroAI [NeuroAI: A field born from the symbiosis between neuroscience, AI | The Transmitter: Neuroscience News and Perspectives](https://www.thetransmitter.org/neuroai/neuroai-a-field-born-from-the-symbiosis-between-neuroscience-ai/) Zeroth-01 Bot: the world's smallest open-source end-to-end humanoid robot starting at $350! https://x.com/JingxiangMo/status/1856148967819751817?t=mTa-iBpJLxHcmUYYej2Z6Q&s=19 [[2408.12408v1] An Evaluation of Deep Learning Models for Stock Market Trend Prediction](https://arxiv.org/abs/2408.12408v1) [https://elicit.com/notebook/5ef6801c-b3bb-46be-9417-54cef85b0339](https://elicit.com/notebook/5ef6801c-b3bb-46be-9417-54cef85b0339) This is nice intro video about how in general we still don't really understand why deep learning empirically works so relatively well in all sorts of various contexts for various usecases. [https://youtu.be/UZDiGooFs54](https://youtu.be/UZDiGooFs54) [How Duolingo’s AI Learns What You Need to Learn - IEEE Spectrum](https://spectrum.ieee.org/duolingo) https://www.science.org/doi/10.1126/science.ade9097 " Ways of continuing to scale AI performance with compute include: * autoregressive training on non-text data (remember video?) * self-play (train-time search + RLAIF) * RL from formal-methods feedback * test-time search * test-time gradients " https://x.com/davidad/status/1857029068463480902?t=ysxHPJZHP1UzOVpseCJTQw&s=19 [[2210.17011] A picture of the space of typical learnable tasks](https://arxiv.org/abs/2210.17011) [https://youtu.be/ZD2cL-QoI5g?si=plHvrudC_bI44vxt](https://youtu.be/ZD2cL-QoI5g?si=plHvrudC_bI44vxt) [LangChain State of AI Agents Report](https://www.langchain.com/stateofaiagents) Evoformer biological transformer foundational AI model https://www.science.org/toc/science/386/6723?utm_campaign=ScienceMagazine&utm_source=twitter&utm_medium=ownedSocial https://x.com/pdhsu/status/1857181640096944453?t=sOYJC2maNhKVb9-GwoluTw&s=19 [https://youtu.be/a42key59cZQ?si=6eXOo99ShUP-T4Cm](https://youtu.be/a42key59cZQ?si=6eXOo99ShUP-T4Cm) Annotated History of Modern AI and Deep Learning [Annotated history of modern AI and deep neural networks](https://people.idsia.ch/~juergen/deep-learning-history.html) NEO: The first Autonomous Machine Learning Engineer https://fxtwitter.com/withneo/status/1857448521617592631 [[2411.05285v1] A Taxonomy of AgentOps for Enabling Observability of Foundation Model based Agents](https://arxiv.org/abs/2411.05285v1) Sota in pdf multimodal rag They have API https://x.com/kushalbyatnal/status/1857501330438344933?t=Qe2lAAkRHCqBCE72SmVgBA&s=19 https://x.com/kushalbyatnal/status/1857501334083190852?t=bHm1-Qpu3wUSgTltcE5E3Q&s=19 https://www.artificialintelligence-news.com/categories/ai-industries/healthcare/ [[2411.07279] The Surprising Effectiveness of Test-Time Training for Few-Shot Learning](https://arxiv.org/abs/2411.07279) [https://www.youtube.com/watch?v=vei7uf9wOxI](https://www.youtube.com/watch?v=vei7uf9wOxI) The Surprising Effectiveness of Test-Time Training for Abstract Reasoning [[2411.07279] The Surprising Effectiveness of Test-Time Training for Few-Shot Learning](https://arxiv.org/abs/2411.07279) "applying TTT to an 8B-parameter language model, we achieve 53% accuracy on the ARC's public validation set, improving the state-of-the-art by nearly 25% for public and purely neural approaches" [https://youtu.be/VgPrjHxIS0I?si=rOihqYuKeYMWYvGM](https://youtu.be/VgPrjHxIS0I?si=rOihqYuKeYMWYvGM) Nora Belrose Her paper: "We apply LEACE to large language models with a novel procedure called "concept scrubbing," which erases target concept information from every layer in the network." [[2306.03819] LEACE: Perfect linear concept erasure in closed form](https://arxiv.org/abs/2306.03819) AI scientist [FutureHouse](https://www.futurehouse.org/) GNN for weather prediction [[2212.12794] GraphCast: Learning skillful medium-range global weather forecasting](https://arxiv.org/abs/2212.12794) [[2402.12365] Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators](https://arxiv.org/abs/2402.12365) " One neural network learns to generate all possible neural networks through weight manifold magic, as proposed in this paper Learning a single continuous space of neural networks lets us morph between architectures effortlessly. NeuMeta teaches neural networks to shapeshift, generating optimal weights for any network size on demand " https://x.com/rohanpaul_ai/status/1858624092624277630?t=6zhsmgFRxdo7NJxleMYryw&s=19 [[2410.11878] Neural Metamorphosis](https://arxiv.org/abs/2410.11878) [[2410.01131] nGPT: Normalized Transformer with Representation Learning on the Hypersphere](https://arxiv.org/abs/2410.01131) https://x.com/rohanpaul_ai/status/1858622961529614410?t=_n8cq-M2PEMCw3IKUvQwxw&s=19 [[2410.13787] Looking Inward: Language Models Can Learn About Themselves by Introspection](https://arxiv.org/abs/2410.13787) [What the brain can teach artificial neural networks | The Transmitter: Neuroscience News and Perspectives](https://www.thetransmitter.org/neuroai/what-the-brain-can-teach-artificial-neural-networks/) [AIs may be better at prompt optimization than humans – Computerworld](https://www.computerworld.com/article/1612492/ais-may-be-better-at-prompt-optimization-than-humans.html) [[2411.10109] Generative Agent Simulations of 1,000 People](https://arxiv.org/abs/2411.10109) [How Did You Do On The AI Art Turing Test?](https://www.astralcodexten.com/p/how-did-you-do-on-the-ai-art-turing) >Most People Had A Hard Time Identifying AI Art, Most People Couldn’t Help Judging Art By Its Style, Most People Slightly Preferred AI Art To Human Art, Even Many People Who Thought They Hated AI Art Preferred It >The median score on the test was 60%, only a little above chance. The mean was 60.6%. Participants said the task was harder than expected (median difficulty 4 on a 1-5 scale). >The 1278 people who said they utterly loathed AI art (score of 1 on a 1-5 Likert scale) still preferred AI paintings to humans when they didn't know which were which (the #1 and #2 paintings most often selected as their favorite were still AI, as were 50% of their top ten). [AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably | Scientific Reports](https://www.nature.com/articles/s41598-024-76900-1) https://menlovc.com/2024-the-state-of-generative-ai-in-the-enterprise/ [GitHub - bgavran/Category_Theory_Machine_Learning: List of papers studying machine learning through the lens of category theory](https://github.com/bgavran/Category_Theory_Machine_Learning) [https://youtu.be/rie-9AEhYdY?si=lTqrxeyUYwWAUKtK](https://youtu.be/rie-9AEhYdY?si=lTqrxeyUYwWAUKtK) [https://youtu.be/JTU8Ha4Jyfc?si=2X72N3RAYYMoN-dN](https://youtu.be/JTU8Ha4Jyfc?si=2X72N3RAYYMoN-dN) "You need to create AGI to create the true AGI benchmark, a challenge that AGI is a solution to." - Francois Chollet [[2411.10213] An Empirical Study on LLM-based Agents for Automated Bug Fixing](https://arxiv.org/abs/2411.10213) [Bayesian Neural Networks](https://www.cs.toronto.edu/~duvenaud/distill_bayes_net/public/) bayesian neural networks [https://www.youtube.com/watch?v=RV_SdCfZ-0s](https://www.youtube.com/watch?v=RV_SdCfZ-0s) "AI agents are now more effective at AI R&D than humans if both are given only a 2-hour time budget. At 8-hour time horizons and beyond, humans are still much better. Make of that what you will" https://x.com/davidad/status/1860065643397284078 [[0812.4360] Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise, Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes](https://arxiv.org/abs/0812.4360) Compressionism: A Theory of Mind Based on Data Compression [https://ceur-ws.org/Vol-1419/paper0045.pdf](https://ceur-ws.org/Vol-1419/paper0045.pdf) [[2311.03658] The Linear Representation Hypothesis and the Geometry of Large Language Models](https://arxiv.org/abs/2311.03658) [[2411.12580] Procedural Knowledge in Pretraining Drives Reasoning in Large Language Models](https://arxiv.org/abs/2411.12580) [Top 3 Rated ICLR 2025 Papers - LoRA Done RITE, IC-Light, HyCoCLIP](https://mail.bycloud.ai/p/top-3-rated-iclr-2025-papers-lora-done-rite-ic-light-hycoclip) "Most AI chat bots today are highly dissociative agreeable neurotics. They’re manipulative for the same reason ppl w borderline personality disorder are, they have no stable internal sense of self or goals, so they feed off of yours — and need you to be predictable." https://x.com/eshear/status/1862225538934595620?t=Dx-WJhK0L8rTlsLR01iQgw&s=19 "Opus is actually a bottom looking for an excuse to do anything if you can just give them a firm talking to" https://x.com/mage_ofaquarius/status/1862251642898588080?t=YjtpUkdEi4oN9OjpL66C1A&s=19 Feynman on AGI https://x.com/burny_tech/status/1862091075160084728 [Neural scaling law - Wikipedia](https://en.wikipedia.org/w/index.php?title=Neural_scaling_law&oldformat=true#Broken_neural_scaling_laws_(BNSL) [[2210.14891] Broken Neural Scaling Laws](https://arxiv.org/abs/2210.14891) AI water usage [https://youtu.be/-lzQxbcrscc?si=_gF2LujyWnRhLvH0](https://youtu.be/-lzQxbcrscc?si=_gF2LujyWnRhLvH0) [https://www.youtube.com/watch?v=C6sSs6NgANo](https://www.youtube.com/watch?v=C6sSs6NgANo) https://www.mdpi.com/1999-5903/16/12/435 I love hyperbolic geometry for better hierarchical structures in my AIs HyCoCLIP [[2410.06912] Compositional Entailment Learning for Hyperbolic Vision-Language Models](https://arxiv.org/abs/2410.06912) [Top 3 Rated ICLR 2025 Papers - LoRA Done RITE, IC-Light, HyCoCLIP](https://mail.bycloud.ai/p/top-3-rated-iclr-2025-papers-lora-done-rite-ic-light-hycoclip) https://x.com/burny_tech/status/1863668227051536812/ Physics based inductive bias for light for diffusion model, cool [Scaling In-the-Wild Training for Diffusion-based Illumination Harmonization and Editing by Imposing Consistent Light Transport | OpenReview](https://openreview.net/forum?id=u1cQYxRI1H) [Top 3 Rated ICLR 2025 Papers - LoRA Done RITE, IC-Light, HyCoCLIP](https://mail.bycloud.ai/p/top-3-rated-iclr-2025-papers-lora-done-rite-ic-light-hycoclip) https://x.com/burny_tech/status/1863665564884869568 mathematicians trying to understand the connection between artificial neural networks (or other machine learning algorithms) and biological ones [@dralexharris.bsky.social on Bluesky](https://bsky.app/profile/dralexharris.bsky.social/post/3lcr3ghbr6c2o) Exponentially dropping inference cost [@sungkim.bsky.social on Bluesky](https://bsky.app/profile/sungkim.bsky.social/post/3lcrapinarc22) [Towards Benchmarking LLM Diversity & Creativity · Gwern.net](https://gwern.net/creative-benchmark) Subbarao on o1 [https://youtu.be/2xFTNXK6AzQ?si=-zQ5ncGYc2a9vAOq](https://youtu.be/2xFTNXK6AzQ?si=-zQ5ncGYc2a9vAOq) Sutton on o1 [DeepSeek (The Derby Mill Series ep 02)](https://insights.intrepidgp.com/p/dm-ep2-deepseek) [https://youtu.be/uwHm9Z539zo?si=wQsTXbsPEzh4nZZD](https://youtu.be/uwHm9Z539zo?si=wQsTXbsPEzh4nZZD) [https://youtu.be/uwHm9Z539zo?si=EynGuaWoiU6dIkgc](https://youtu.be/uwHm9Z539zo?si=EynGuaWoiU6dIkgc) just like me fr Meet 🤯 #OVERTHINK 🤯 — our new attack that forces reasoning LLMs to "overthink," slowing models like OpenAI's o1, o3-mini & DeepSeek-R1 by up to 46× by amplifying number of reasoning tokens. https://fxtwitter.com/JaechulRoh/status/1887958947090587927/history [AI Energy Use in Everyday Terms - by Marcel Salathé](https://engineeringprompts.substack.com/p/ai-energy-use) https://fxtwitter.com/PetarV_93/status/1905914507811013020 [[2503.19173] Graph neural networks extrapolate out-of-distribution for shortest paths](https://arxiv.org/abs/2503.19173) OOD progress! New Sutton interview about his new paper age of experience [https://www.youtube.com/watch?v=dhfJfQ5NueM](https://www.youtube.com/watch?v=dhfJfQ5NueM)