#Research
Curated coverage on «Research»: releases, research, benchmarks, open-source.
Δτ-Bench: Best AI Agent Solves Only 42.3% of Tasks in a Shifting World
Researchers from UCL and co-authors have released Δτ-Bench: 1,588 tasks across 42 domains where tools vanish and get replaced. The best agent managed only 42.3%.
DynaFlow teaches a lead agent to write its own workflows, beats the baseline by 10.9 points
NUS researchers trained a lead agent on Qwen3.5-27B to write its own programs for sub-agents. Result: 88.08% accuracy across six benchmarks, plus 10.90 points over…
SATE-Bench: Multimodal Models Give Up on Scene Exploration Too Soon
A new benchmark, SATE-Bench, covers 316 real-world scenes and 2,525 QA pairs. Top MLLMs stop their scene review prematurely while staying confident in their…
LLM Agents Spent 20 Days Building an AI for Hearthstone Battlegrounds That Beats Top-10 Players
CMU Researchers Teach RL Models to Stop Padding Answers: FORK Lifts Win Rate from 22% to 74%
Mycelium: Researchers Break Down Hallucination Benchmarks Along Three Axes
AV-FABLE: Researchers Assemble 998 Hours of Video to Hunt for Pinpoint Deepfakes
Raspberry Pi 5 off the grid: a signed poisoned knowledge package sailed through every check
The authors ran offline RAG on Qwen2.5 1.5B/3B/7B on a Raspberry Pi 5: signature checks missed 6 of 9 update attacks, while layered verification caught all 9 in…
RoboJev-27B judges robot success from 10 frames: 0.993 AUROC on SO-101
An open 27-billion-parameter VLM watches 10 frames and tells you whether a robot nailed its task: 95.2% balanced accuracy, 0.26 seconds per episode, and about $0.19…
Stanford PhD Derives the Transformer From Scratch, Starting With One Memory Problem
Zhejiang University and Peking University Propose SPADER for Multi-Answer QA
Xiaohongshu and Fudan University Build DR-Rubric for RL on Open-Ended Tasks
DynStream from Beijing builds a dynamic 3D scene from video without camera poses
Sparse Autoencoders: Pairwise Interactions Lift Rank Fidelity to 0.7790
The new SAR metric shows SAE features add up better than a random basis in all 18 settings, but accurate steering still requires pairwise interactions.
Weight editing without retraining cut sexist responses from 57.9% to 24.3%
Researchers at AI Interpretability @ Illinois developed a weight-editing method for models. Without retraining, it cut sexism by 33.6 points and boosted…