Well, the tokenizer I used was trained on large quantities of data — I filtered the tokens based on yet more data from FineWeb. Question is if that's acceptable according to your rules…
MACHINE LEARNING
-
AI Improves Grid Optimization and Predictive Maintenance Reliability
By
–
Some early progress can be seen in grid optimisation and predictive maintenance, where AI is improving reliability, reducing downtime, and creating a more intelligent use of resources.
-
Celebrating shift from JSON tool calls to code blobs
By
–
Finally! Congrats, a step in a great direction!
We've been stuck in this local optimum of "writing actions as JSON blobs with individual tool calls in them" for much too long. Just let models write tool calls in code blobs! -
ThinkMorph: Emergent Properties in Multimodal Chain-of-Thought
By
–
ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning Gu et al.: https://
arxiv.org/abs/2510.27492 #ArtificialIntelligence #DeepLearning #MachineLearning -
Multi-Agent Neural Cellular Automata Develop Complex Emergent Behaviors
By
–
Excited to release our new work: Petri Dish Neural Cellular Automatahttps://t.co/4mWwFOnbYl
— hardmaru (@hardmaru) 5 novembre 2025
We investigate how multi-agent NCAs can develop into artificial life 🦠 exhibiting complex, emergent behaviors like cyclic dynamics, territorial defense, and spontaneous cooperation! https://t.co/Wjp6KkurSQExcited to release our new work: Petri Dish Neural Cellular Automata https://
pub.sakana.ai/pdnca We investigate how multi-agent NCAs can develop into artificial life exhibiting complex, emergent behaviors like cyclic dynamics, territorial defense, and spontaneous cooperation! -
DataRater: Automatic Continuous Learning for Model Training Examples
By
–
DataRater: can you automatically+continuously learn which examples will help the model the most? Fun working on this w/ awesome set of co-authors: @luisa_zintgraf
, @dancalian
, @greg_far
, @iurii_kemae, @matteohessel
, @shar_jeremy
, @junh_oh
, András György, Tom Schaul, @hado
, & -
Scalable and Compounding AI Architecture Benefits
By
–
What's better than a scalable AI architecture? A scalable *and compounding* AI architecture.
-
Autonomous AI Through Dynamic Model Synthesis and Self-Improving Abstractions
By
–
The path to autonomous AI is a system that learns to solve new problems by synthesizing models of them on the fly (as code), and that gets smarter over time by adding new abstractions to its own library (also as code), compounding its capabilities. Not a static map — rather, an
-
AI Research Breakthrough with International Mathematics Experts
By
–
Finally, this work won't be possible with the co-authors: @DawsenHwang
, @HoangT1215
, @g01na2
, @junsukim
, @gjb_ai
, @jon_lee0
, @Swarooprm7
, @hmichalewski
, @XingyouSong
, @thtrieu_
, @quocleix
, @jj_at_brown
, our IMO experts (who account for a total of 10 gold and 5 silver IMO medals), -
GradingBench releases 1000 human-graded IMO proof data points
By
–
Last but not least, #GradingBench has 1000 human grading data points from our IMO effort on the advanced IMO-ProofBench. To ensure a robust evaluation, the dataset has been balanced across 4 simplified grading categories (Correct, Almost, Partial, Incorrect). We released all