#Research
Curated coverage on «Research»: releases, research, benchmarks, open-source.
Terence Tao writes for mathematicians on Lean's reliability and the role of AI
Tao breaks down what mathematicians should and shouldn't trust in Lean as AI increasingly writes proofs. The post scored 53 points on Hacker News in its first hours.
Google DeepMind and Biohub Explain Why AlphaFold Didn't Close the Protein Folding Problem
Pushmeet Kohli of DeepMind and Sal Candido of Biohub break down what AlphaFold left unsolved after the Nobel Prize and 200+ million predicted structures.
Stanford Online Posts Lecture 8 of AA274A on Robot Sensors and Computer Vision
Stanford Online has published a lecture by Marco Pavone from the 2019 course: robot sensors and an introduction to computer vision. The Principles of Robot Autonomy…
LLMs Taught to See Every Letter Find Reversing Words Easier
Researchers Study How Chatbots Solve Problems Without Thinking in Symbols Like Humans Do
PANTHER: AIRI student project wins FLARE 2026, finding tumors in 11 seconds
YOLO Annotator: An Offline Labeling Tool for bbox, OBB and Panoptic Segmentation Under GPL-3.0
OpenAI claims an AI-generated proof of the Navier–Stokes problem, and it is hard to verify
Bloomberg reports OpenAI unveiled an AI proof of the Navier–Stokes problem a month ago. MIT's Justin Solomon explains why even experts struggle to read it.
Aether AI shows a manipulator with a causal core in a seven-minute video full of interference
Aether AI released a seven-minute video of a manipulator working with a coffee machine and a microwave while people interfere at every step. The company is betting…
Gordon Brown Calls for Innovation to Become the Centerpiece of British Policy: £5,000 Per Family
8B RL-Trained Agent Teaches a Video Model to Look Up Data and Compute Physics
OpenAI Publishes 722 Math Papers Written by AI
Video Generation With an Expression Gallery Beats Baseline Models on Identity Similarity
Fluid-Gen-Zero hands water physics to a simulator and cuts fluid flow error by 84%
A training-free framework links a physics simulator to CogVideoX- and Wan-based video models, cutting object trajectory error by up to 81.5% and fluid flow error by…
LVSPM Beats VGGT at Camera Pose Estimation on 16–256 Frames and Doesn't Break Down on Large Scenes
The new LVSPM model recovers camera poses and synthesizes novel views from hundreds of uncalibrated images at once, outperforming VGGT on RealEstate10k, Co3Dv2 and…