#Research
Curated coverage on «Research»: releases, research, benchmarks, open-source.
Wits Physicists Build a "Light Brain" That Reads Hidden Patterns in Light
Researchers at the University of the Witwatersrand built an optical system that recognizes hidden information in complex light patterns. The goal is faster…
Google DeepMind Has a Staff Philosopher: Henry Shevlin Studies AI Ethics at 80% Time
Cambridge philosopher Henry Shevlin has spent 80% of his time at Google DeepMind since May 2026. He explains why AI ethics can't be studied from an ivory tower.
Unitree G1 Humanoid Catches Boxes of Up to 2 kg Thrown From 7 Meters Using Reflex
Researchers from UW, USC, NUS, UNC and Allen AI taught a Unitree G1 to catch thrown boxes weighing up to 2 kg from up to 7 meters away, using only an RGB-D camera…
Jiuzhang Yunji Enters Tao's Math Challenge After Letter From 28 Fields Medalists
New arXiv Paper Claims AI Models Can Provably Hide Any Capability
Tsinghua Team Releases OSWorld-Science: Testing AI Agents in Scientific Software
UniWAM from 欧拉万象 and HKUST(GZ) Combines World Understanding, Prediction and Robot Action in One Model
Apple proposes SelfCompact: the agent decides for itself when to compress context
Apple researchers removed the fixed token threshold from context compression: in SelfCompact, the model chooses when and how to compact its context through a…
Apple tested LLMs as translation judges: MQM and ESA are hard for both humans and models
Apple ML Research compared LLM and human quality annotations for machine translation under the MQM and ESA schemes. The title gives the verdict: both sides struggle.
DisaggKD moves the teacher onto separate nodes and speeds up distillation of Qwen3.5-35B-A3B into 9B
Copy-Paste Language Models Add a Copy Head to LLMs With 135M–1.7B Parameters
LeanPRM: Lean Labels 152,000 Proof Steps Without Humans and Lifts Accuracy to 74.7%
Researchers audited 6,309 agent skills: half have shell access
Microsoft Research Unveils StudentSim, a Student Simulator for AI Tutors
Microsoft Research showed StudentSim, a system that simulates student behavior to test AI tutors in chess, writing and math before trials with real people.
Researcher proposes four rungs of meaning for AI, from nouns to pacts
A 12,000-word AI Alignment Forum post asks how agents with different world models can agree on rules. The author builds a four-step ladder and offers bounties for…