Jake will be working with a lot of folks on continual learning – if that’s interesting, reach out to him!
RESEARCH
-
Policy Brief: Monitoring AI in FDA-Authorized Medical Devices
By
–
New policy brief: 76% of FDA-authorized medical AI devices are radiology tools — yet most lack robust post-deployment monitoring. Our latest brief introduces a practical framework for real-time performance oversight of clinical AI. https://
hai.stanford.edu/policy/operati
onalizing-real-time-monitoring-of-clinical-ai
… -
OrthoReg: A New Method for Weight Disentanglement in Fine-Tuning
By
–
What makes task arithmetic actually work? Researchers from Nanjing University, U. Wollongong, and NTU Singapore introduce OrthoReg: a simple method that enforces orthogonality in weight updates during fine-tuning. This promotes weight disentanglement and consistently boosts
-
Standardized Blueprints Needed for AI Organizations
By
–
We will need shareable and standardised Blueprints for AI Organisations BLUEPRINT.MD
-
LangChain Labs launches focused on continual learning
By
–
JUST IN: We're launching LangChain Labs. A new applied research effort focused on Continual Learning.
-
GPT-5.5 solves Erdős problems, post-AGI research feels parallel
By
–
GPT-5.5 has a certain magic about it. It solves one Erdős problem after another. this is what post-AGI research may actually feel like. Not one dramatic "AI solves math" moment, but dozens of parallel discoveries, anonymous contributors, formal proofs as trust infrastructure,
-
Building AI Products with Open-Source Models and Sovereign Infrastructure
By
–
Sydney showed up Huge thanks to the founders, devs, builders, & partners who joined us for SambaHouse with Equinix & SCX. Great convos, live demos, & a room full of people building real AI products with open-source models & sovereign infrastructure.
-
Multimodal AI outperforms physicians in simulated study
By
–
A multimodal AI (AIME) had superior performance compared with 18 physicians for almost every metric (29 of 32 axes). Randomized, but simulated, not real-world clinical practice. @GoogleDeepMind @alan_karthi @RyutaroTanno https://
nature.com/articles/s4159
1-026-04371-0
… -
New generative model research uses low-rank Nyström approximation for faster training
By
–
Ali did some amazing work on the hottest new generative model: drifting models (from Mingyang Deng @Goodeat258 et al., out of Kaiming He's group).
— Gautam Kamath (@thegautamkamath) 14 mai 2026
Speeds up training a lot using low rank Nyström approximation. Check out Ali's full thread. Paper and code available! https://t.co/MDSo47g4rUAli did some amazing work on the hottest new generative model: drifting models (from Mingyang Deng @Goodeat258 et al., out of Kaiming He's group). Speeds up training a lot using low rank Nyström approximation. Check out Ali's full thread. Paper and code available!
-
Reflections on the development and training of video AI models
By
–
It's heavy how strong the blow was to the Seedance 2.0 table that completely halted the cadence of video models we'd been carrying up to then. May they rest in peace, all those model checkpoints that were trained and never published because they weren't up to par