Diffusion seems like a much more natural way to get good reasoning to me. And you can directly scale as many more steps in diffusion as you want – no problem!
MACHINE LEARNING
-
Google DeepMind partners with Fenris Creations for Eve Online AI research
By
–
I've always been passionate about games and they've played a big part in @GoogleDeepMind
’s history, as the perfect proving ground for AI. Thrilled to announce this research partnership with @FenrisCreations – @EveOnline is one of the most extraordinary games ever built and has an -
Agent Observability: Storing Feedback with Traces for Learning Systems
By
–
Hot take, the models are not going to converge in the way that most folks talk about
-
Elon Musk’s X Agents Learn from User Posts
By
–
And now I know why @elonmusk paid $45 billion for X. His agents are getting smarter at a faster rate and are even more expensive. 🙂 You didn't know you were teaching agents by posting on X, did you? 🙂
-
AI Model Analogy: Hermes vs. OpenClaw
By
–
I LOVE @garrytan
's analogy: "
@NousResearch
's Hermes is a beautiful brand new reliable Honda and @OpenClaw is a Ferrari that you need to bring a wrench for when it breaks down on the side of the road" -
Fully Homomorphic Encryption: Breakthroughs in Speed and Adoption
By
–
Fully homomorphic encryption was invented in the 1980s.
— Cerebras (@cerebras) 6 mai 2026
Why wasn't it adopted sooner? A 100,000x slowdown, driven by memory boundedness. Ajay Joshi from @CipherSonicAI explains how his team got it down to less than 2x.
(if this pattern sounds familar… LLM inference is… pic.twitter.com/gBVRMCBhBRFully homomorphic encryption was invented in the 1980s. Why wasn't it adopted sooner? A 100,000x slowdown, driven by memory boundedness. Ajay Joshi from @CipherSonicAI explains how his team got it down to less than 2x. (if this pattern sounds familar… LLM inference is
-
Microsoft Research Paper on Agent-Based Interpretability for AI
By
–
NEW paper from Microsoft Research. (bookmark it) The entire interpretability literature is built around human readers. As more analysis gets delegated to agents, the right target of interpretability shifts. This paper is a recipe for designing tools that agents can actually
-
Device with Sensors to Interpret Emotions
By
–
This is just a total lack of creativity on behalf of educators. Years ago when I was going to college my chemistry professor said, on her first day in class, "you are allowed to cheat in my class." "You can bring the books to my chemistry exams." "You can bring a
-
AI building emotional friends with composable graph-based memory and RL environments
By
–
This could actually be a very strong bear case for Tesla the company. If their cars are so good that they really don't need any maintenance for 100s of thousands of miles, the whole upgrade cycle could be well below the industry average.