Even better: Ordinal regression (/classification) Shameless plug:
MACHINE LEARNING
-
Tinker API: Accessible GPU Infrastructure for ML Research
By
–
GPUs are expensive and setting up the infrastructure to make GPUs work for you properly is complex, making experimentation on cutting-edge models challenging for researchers and ML practitioners. Providing high quality research tooling is one of the most effective ways to improve research productivity of the wider community and Tinker API is one step towards our mission there. Tinker API is built on top of our experimental results on fine-tuning with LoRA: thinkingmachines.ai/blog/lor… Beta starts and you can join the waitlist today: thinkingmachines.ai/tinker/
→ View original post on X — @lilianweng, 2025-10-01 18:29 UTC
-
AI Models Predict Alzheimer’s and Parkinson’s Risk Progression
By
–
What if an AI model could predict your risk and progression of Alzheimer’s or Parkinson's? Scholars are building a world model of the brain that could create better predictions for diagnosis and care. Learn more at the Hoffman-Yee Symposium on Oct. 14: https://t.co/rbgxicx1sy pic.twitter.com/dxIdav9pJq
— Stanford HAI (@StanfordHAI) 1 octobre 2025What if an AI model could predict your risk and progression of Alzheimer’s or Parkinson's? Scholars are building a world model of the brain that could create better predictions for diagnosis and care. Learn more at the Hoffman-Yee Symposium on Oct. 14: https://
hai.stanford.edu/events/hoffman
-yee-symposium-2025
… -
Moonlake Launches AI Platform for Real-Time Simulations and Games
By
–
Moonlake is launching publicly today, focused on generating simulations and games. Simulations are an important application of generative models that I’ve advocated for since 2014 https://t.co/Qti9LM9wul
— Ian Goodfellow (@goodfellow_ian) 1 octobre 2025Moonlake is launching publicly today, focused on generating simulations and games. Simulations are an important application of generative models that I’ve advocated for since 2014 Moonlake (@moonlake) We raised $28M seed from Threshold Ventures, AIX Ventures, and NVentures (Nvidia's venture capital arm) —alongside 10+ unicorn founders and top AI researchers— to build reasoning models that generate real-time simulations and games. Models are bottlenecked by practical simulations that can act as Reinforcement Learning environments. Human self-expression is bounded by tools that let us create alternate realities. At Moonlake, we are building a future where anyone can create interactive worlds, bring their child-like wonder to life, learn within them, and most importantly, share experiences with people we care about. More in 🧵 — https://nitter.net/moonlake/status/1973432805519708457#m
→ View original post on X — @goodfellow_ian, 2025-10-01 17:54 UTC
-
Sutton’s Bitter Lesson Shapes LLM Research Approaches
By
–
Finally had a chance to listen through this pod with Sutton, which was interesting and amusing.
— Andrej Karpathy (@karpathy) 1 octobre 2025
As background, Sutton's "The Bitter Lesson" has become a bit of biblical text in frontier LLM circles. Researchers routinely talk about and ask whether this or that approach or idea… https://t.co/EQ4snNmiTMFinally had a chance to listen through this pod with Sutton, which was interesting and amusing. As background, Sutton's "The Bitter Lesson" has become a bit of biblical text in frontier LLM circles. Researchers routinely talk about and ask whether this or that approach or idea
-
World Models Learned from Video: New Evidence and Implications
By
–
More evidence that world models can be learned from video alone (and a small amount of video at that): https://
arxiv.org/pdf/2509.24527 -
Google Trains Single-GPU AI Agent to Mine Minecraft Diamonds
By
–
Fast progress in training AI agents to interact with the world. Training on just 2,541 hours of Minecraft video, Google built an AI that runs on a single GPU & was able to mine diamonds offline (which takes an average of 24,000 clicks). The same approach may work for AI robots.
-
Model Achieves Faster Inference Speed Than Smaller Competitors
By
–
But the best feature is inference speed: it's significantly faster than models that are much smaller.
-
LFM2-1.2B Extension: Novel Audio Foundation Model Architecture
By
–
It has a super interesting architecture that extends LFM2-1.2B.
— Maxime Labonne @ ICLR (@maximelabonne) 1 octobre 2025
Unlike other audio foundation models, we disentangle the representations of audio input and output.
More info in the blog post. pic.twitter.com/cei9MSoCRuIt has a super interesting architecture that extends LFM2-1.2B. Unlike other audio foundation models, we disentangle the representations of audio input and output. More info in the blog post.
-
Present Your AI and Data Solutions to Large Audience
By
–
Here is your chance to present your best AI and data ideas to a large audience. http://
2.sas.com/6019AZIpV Maybe you are engineering AI solutions using large language models (LLMs), synthetic data generation or digital twins. We’re looking for these presentations and more, to