Cool, now it's possible to generate high-fidelity AI images in just two steps! Enter ArcFlow from Fudan University and Microsoft Research Asia. While most tools use straight-line shortcuts that sacrifice quality, ArcFlow uses a non-linear approach that follows the natural,
RESEARCH
-
AI Reshapes Internet Infrastructure at MWC 2026
By
–
The Internet is about to change — not because of what AI can do, but because of what AI does to the network itself. #Sponsored I'm at MWC 2026 in Barcelona at the Nokia stand, and the research from Nokia Bell Labs stops you in your tracks. Right now, as you read this, AI is
-
OpenAI GPT-5.4: Beyond Scale in AI Race
By
–
Everyone keeps saying the AI race is about bigger models. More parameters.
More data.
More compute. That has been the playbook for the last two years. But when I looked at OpenAI’s new release this week, one detail caught my attention. OpenAI introduced 𝐆𝐏𝐓-𝟓.𝟒, its -
Introduction to LLM Post-Training Techniques Presentation
By
–
A big thank you to @itsmaddox_j for the invitation, and to professors @niclane7 and @contactrika for having me! slideshare.net/slideshow/int…
→ View original post on X — @maximelabonne, 2026-03-06 10:46 UTC
-
53 Slides on Post-Training Algorithms and Data Quality
By
–
I'm releasing 53 slides on post-training, covering core algorithms like DPO and GRPO, as well as data quality, synthetic data pipelines, and on-policy training. I had the pleasure of presenting it yesterday as a guest lecturer in Cambridge, UK
→ View original post on X — @maximelabonne, 2026-03-06 10:45 UTC
-
Models Know Longer Isn’t Always Better
By
–
the deeper point here connects to something the field keeps rediscovering. we trained reasoning models to think longer. then we discovered longer doesn't mean better. now this paper shows the models themselves already know that. they're generating stop signals that our inference
-
DeepSeek and Qwen3 performance improvements
By
–
specific numbers worth sitting with: > DeepSeek-R1-7B on MATH-500: 93% accuracy (up from 91.6%), tokens cut from 3,871 to 2,141 > DeepSeek-R1-1.5B on AIME 2025: accuracy jumps 6.2 percentage points > Qwen3-8B: response length halved from 18,342 to 9,183 tokens with no accuracy
-
SAGE-RL: Simple RL Training Modification
By
–
then they fold it into training with SAGE-RL. dead simple modification: in standard reinforcement learning (GRPO), you sample 8 responses per question. SAGE-RL replaces 2 of those 8 with SAGE-generated samples. the other 6 stay normal. one-line code change. the model learns to
-
SAGE: Efficient Reasoning with Confidence Checks
By
–
their solution: SAGE (Self-Aware Guided Efficient Reasoning). instead of generating token by token, SAGE extends chains in whole reasoning steps. after each step, it checks: is the model confidently signaling it wants to stop? if yes, reasoning ends. no fine-tuning. no new
-
Researchers test AI self-awareness in reasoning
By
–
here's where it gets interesting. the researchers probed whether models internally "know" they're done. they introduced TSearch, which scores partial reasoning traces by cumulative log-probability across the entire chain, not just the next token. when you let the model explore