// Efficient Evolution of Web Agents // Web agents waste a lot of compute on cyclic reasoning loops and unproductive exploration. This new research introduces WebClipper, a framework that models web agent search processes as state graphs and prunes them into minimal directed
MACHINE LEARNING
-
Alibaba releases Qwen 3.5, analyzed via Blevlabs on X
By
–
The Qwen Report. Today @AlibabaGroup released Qwen 3.5. A major new multimodal model. So I had @blevlabs analyze all posts here on X and a few other places and write up this report. All built via the X API. This is an example of the value that's here on X that very few see.
-
Building an AI-first reading experience for X
By
–
Hah. I am building a new kind of X reading experience all read by AI first. 🙂
-
Paper and Code: Youtu-Agent Training Free GRPO Implementation
By
–
Paper: https://
arxiv.org/abs/2510.08191
v1
…
Code: https://
github.com/TencentCloudAD
P/youtu-agent/tree/training_free_GRPO
… -
Critique of Machine Learning Research for Tabular Data
By
–
Every ML for tabular data paper I've ever come across.
-
Which exponential trends are you riding right now?
By
–
Which exporential are you trying to ride right now?
-
How Recent Physics Result with AI Came Together
By
–
how our recent physics result with AI came to be
-
Categorical Flow Maps: Fast Discrete Diffusion for Single-Step Generation
By
–
Great work by @FEijkelboom et al Floor Eijkelboom (@FEijkelboom) Discrete diffusion — but fast? ⚡️ Test-time inference — but for discrete data? 🧠 Categorical Flow Maps: continuous transport toward the simplex, turning discrete generation into a single-step problem. Built on Variational FM (CatFlow), we obtain (self-)distillation from scratch. Language, molecules, and test-time steering — one framework. Scaling to LLMs and foundation models next. Watch this space 👀 — https://nitter.net/FEijkelboom/status/2023412552294686734#m
→ View original post on X — @wellingmax, 2026-02-16 15:38 UTC
-
Vision Redefined: From Representations to Perception-Action Loops
By
–
In my recent blog post, I argue that "vision" is only well-defined as part of perception-action loops, and that the conventional view of computer vision – mapping imagery to intermediate representations (3D, flow, segmentation…) is about to go away. vincentsitzmann.com/blog/bit…
→ View original post on X — @shiqi_yang_147, 2026-02-16 15:31 UTC
-
LLM Agents Struggle With Multi-Step Scientific Tool Use
By
–
On evaluating multi-step scientific tool use in LLM agents. SciAgentGym provides an interactive environment with 1,780 specialized tools across 4 scientific disciplines. The core finding: even advanced models like GPT-5 see success rates drop sharply from 60.6% to 30.9% as