Oh yeah you probably will. I am currently running tons of experiments to decide which of the Olmo 3 and DeepSeek V3.2 GRPO tweak I should add in the final version.
No side objective really except I'd do an honest evaluation to make sure that the model actually learns well.
MACHINE LEARNING
-
Experimenting with Olmo 3 and DeepSeek V3.2 GRPO tweaks
By
–
-
Model Convergence Shifts AI Differentiation to Data Foundations
By
–
On @TBPN
, Databricks CEO and co-founder @alighodsi joined @johncoogan and @jordihays to discuss how model capabilities are converging, and why differentiation increasingly comes from strong data foundations and reasoning over proprietary data. They also talked about how -
Differentiable Programming: Representation vs Learnability Challenge
By
–
The ability to embed your discrete function on a differentiable curve is a prerequisite for differentiable programming, but it is *far* from sufficient, the hard part is getting GD to find the right function Anything can be represented, not everything can be learned
-
OpenAI Prepares Codex Upgrade with GPT-5.2-Codex-Max
By
–
BREAKING : OpenAI is preparing to upgrade Codex, likely with GPT-5.2-Codex-Max. One last leap this year? Caribou
-
New Toolkit for Extending Training Data to Different Robot Embodiments
By
–
Excited to release this toolkit for extending training data to different robot embodiments: https://t.co/sqhKr7akE8
— Ken Goldberg (@Ken_Goldberg) 17 décembre 2025Excited to release this toolkit for extending training data to different robot embodiments:
-
Gradient Descent Limitations in Discrete Reasoning Spaces
By
–
Gradient descent is a powerful tool for optimization spaces that verify the manifold hypothesis, but the space of reasoning is discrete and combinatorial. GD fails in cliff-like landscapes where a single discrete change (a logical step) alters the entire outcome. Unless…???
-
General Intelligence Emerges from Adaptation to Novel Situations
By
–
What's remarkable is that general intelligence emerges evolutionarily from the simple goal "you need to survive/succeed through a series of ever-novel, often adversarial situations" General intelligence is a situated process of efficient adaptation to novelty
-
Universal Intelligence and Scientific Understanding of Natural Laws
By
–
Because our universe follows stable laws, a sufficiently general intelligent system adapted to it, like human-driven science, can eventually model any phenomenon within it. Human intelligence may not be "universal" in the mathematical sense (see No Free Lunch theorem), but we
-
Small Model Outperforms DeepSeek-r1 With Recursive Reasoning
By
–
Recursive reasoning beats multi-billion-parameter models You can now easily train your own 7M param model from scratch and outperform DeepSeek-r1 on ARC-AGI 1 We provide a simple speedrun script that handles setup, training, and eval in one go.
-
Open-source AI paper implementation with pretrained weights released
By
–
Checkout our adaptation of @jm_alexia
's work here: https://
github.com/alphaXiv/paper
-implementations
… We also provide pretrained weights for Sudoku, Maze, and ARC-AGI for easy evaluation!