Things we’re thankful for… – that someone figured out how to program sand
– that Geoffrey Hinton didn’t become a carpenter in his 20s
– that Alan Turing’s mom met Alan Turing’s dad
– that Marvin Minsky wasn’t always right
– that attention is all you need
– that “Move 37” wasn’t
INNOVATION
-
Groq’s Thanksgiving: Gratitude for AI and Computing History
By
–
-
Maya Man’s StarQuest: AI art project exploring imitation learning and dance
By
–
Artist Maya Man describing her fantastic new art project StarQuest, a meditation on AI and imitation learning (based on her personal childhood experience with competitive dance):
-
Google Prepares Proactive Mode for Jules SWE Agent
By
–
Google is working on Scheduled tasks for Jules SWE Agent and preparing "Proactive Mode" for the release.
— 🚨 AI News | TestingCatalog (@testingcatalog) 27 novembre 2025
Proactive mode was teased earlier and will enable Jules to proactively scan the codebase and fix bugs. pic.twitter.com/HncKxk6kLmGoogle is working on Scheduled tasks for Jules SWE Agent and preparing "Proactive Mode" for the release. Proactive mode was teased earlier and will enable Jules to proactively scan the codebase and fix bugs.
-

DeepSeek Math-V2 Gold IMO Model Now Open Source
By
–
As far as I know, there isn't any chatbot or API that gives you access to an IMO 2025 gold-medalist model. Not only does this change today, but you get to download the weights with the Apache 2.0 open-source release of @deepseek_ai Math-V2 on @huggingface
! Imagine owning the -
Latent Process Planning: 6K Parameters Matching 100M Model Capability
By
–
A latent process that operates over plans? I've been working on this recently! What's most fascinating to me: my 6k parameter system can match aspects of models 100,000x bigger. Scaling laws apply very differently too…
-

MLflow GenAI Evaluation Token-Based Pricing Reduces Costs
By
–
As AI agents move from prototype to production, success depends on accuracy — and accuracy depends on whether the agent understands your domain. Evaluation is critical. That’s why we recently introduced token-based pricing for MLflow GenAI evaluation, reducing evaluation costs
-

NVIDIA Research: Efficient AI Beyond Model Scaling
By
–
Banger paper from NVIDIA. Bigger models aren't always the answer. However, the default approach to improving AI systems today remains scaling up. More parameters, more compute, more cost. But many tasks don't require the full power of a massive model. This new research
-
Queue Position 2310: Start Building Apps in Simulation Now
By
–
Got it, you're #2310 in the queue. Should be shipped before Christmas! In the meantime, you can start building apps in simulation:
-
Ilya Sutskever Plans Near-Term AI Releases Over Super Intelligence
By
–
I can't believe I haven't seen the most obvious take from the @dwarkesh_sp interview with @ilyasut – Ilya will ship something soon, instead of waiting for super intelligence. Direct quote below & couple of other clips hinting to that.
— Peter Gostev (@petergostev) 27 novembre 2025
Ilya: "One of the ways in which I’ve changed… pic.twitter.com/JaSNFu15DyI can't believe I haven't seen the most obvious take from the @dwarkesh_sp interview with @ilyasut – Ilya will ship something soon, instead of waiting for super intelligence. Direct quote below & couple of other clips hinting to that. Ilya: "One of the ways in which I’ve changed
-
Open-Source AI Collaboration: How Constraints Drive Global Innovation
By
–
At the #BloombergNewEconomy Forum in Singapore, I discussed the future of AI, and why open-source is how rest of the world can work together to develop AI. When you don’t have hundreds of billions of dollars of funding, you’re forced to collaborate.
— hardmaru (@hardmaru) 27 novembre 2025
Constraints breed innovation. pic.twitter.com/YEvpvA9mNDAt the #BloombergNewEconomy Forum in Singapore, I discussed the future of AI, and why open-source is how rest of the world can work together to develop AI. When you don’t have hundreds of billions of dollars of funding, you’re forced to collaborate. Constraints breed innovation.