I am once again asking for your support of DoRA. Friends don't let friends use LoRA. Use DoRA and make 2026 a good year.
MACHINE LEARNING
-
AI engines build mind-blowing software, but search engine creation is hard
By
–
Software engineers are already telling me that their AI engines, like Anthropic's Claude Opus or ChatGPT's advanced models, are building software so advanced that it would blow your mind. So, just build your own search engine. Well, except that's not really possible that easily
-
Grok in new Teslas enables voice-controlled driving
By
–
Grok is already in new Teslas. It's really amazing—you talk to it and tell it what you want to do, and it takes you there
-
Projected use of imitation learning in future robot combat leagues
By
–
Bon, je pense qu’en 2026, on va voir apparaître les premières ligues de combat de robots.
— Defend Intelligence (Anis Ayari) (@DFintelligence) 1 janvier 2026
Au début, ils seront télécommandés, puis sûrement avec un mode d’apprentissage basé sur l’imitation learning.
Qui a 5 millions d’euros ? On lance une ligue en France.
Les combats de… pic.twitter.com/XSwJ6KWF1sBon, je pense qu’en 2026, on va voir apparaître les premières ligues de combat de robots. Au début, ils seront télécommandés, puis sûrement avec un mode d’apprentissage basé sur l’imitation learning. Qui a 5 millions d’euros ? On lance une ligue en France. Les combats de
-
Publishing Barriers for Experimental AI Research Observations
By
–
Thanks for sharing – very well said 🙂
It's really sad that experimental observations have always been nearly impossible to publish in AI/ML journals and conferences, and has definitely held back the field. -
DexDrummer: Multi-Fingered Robot Hands Dexterous Drumming Task
By
–
Interesting new task for multi-fingered robot hands! : DexDrummer: In-Hand, Contact-Rich, and Long-Horizon Dexterous Robot Drumming – https://
goo.gl/scholar/xdc9gb -
DeepSeek’s 2026 hyper-connection breakthrough
By
–
The whale is back DeepSeek dropping manifold-projected hyper-connections right after the holidays is the kind of energy we needed for 2026. Stabilizing those long-range skips by enforcing geometric alignment instead of letting them warp the representation space elegant fix
-
Systematic AI Training Pipeline Evolution
By
–
The training pipeline is genius: Stage 1: CPT on 500B tokens of structured code tasks
Stage 2: Two-stage SFT with error masking
Stage 3: RL with chunk-level optimization Each stage builds on the last. No shortcuts. Just systematic capability building. -
IPA: Chunk-Level Credit Assignment Breakthrough
By
–
Here's the breakthrough nobody expected: They introduced IPA (Interaction-Perceptive Agentic Policy Optimization) that assigns credit at the CHUNK level, not token level. Tokens are too fine. Trajectories are too coarse. Chunks align with actual tool-use semantics.