2026
10,780 articles · 136 days · 10 categories · avg 79 per day
CoRE Teaches AI to Choose Its Own Knowledge-Base Retrieval Rules Without Labeled Data
CoRE jointly trains a rule router and a retriever without query-rule pairs. It was tested on three STaRK-QA domains: scientific discovery, biomedicine and…
OpenAI shows how an engineer's dot triaged a night-time incident and launched Codex
In an OpenAI video, engineer Ali hands an on-call incident to his dot at 3 a.m. The dot opens a thread in Codex, and Ali steers the fixes from his phone through…
Robots learn from 20 854 hours of video: how Sharpa is building dexterous manipulation
Robotics data is moving from costly teleoperation to human video: EgoScale used 20 854 hours. Here is how Sharpa fits its system into that logic.
Synchronous MoE decoding makes GPUs wait for the busiest group
A Company from Nanjing University's Orbit Bets on a Fourth Driver of AI Growth
MTEdit: 500 thousand editing trajectories teach models not to lose quality over 8 steps
Xiaomi-Lens-0, a 4B model, beats Qwen3.5 by 5.6 points on RefCOCO
OfficeTown Simulates 23 Organizations and Trains Agents for Office Work on 406 Tasks
Researchers built OfficeTown, a generative simulation of offices that produced a benchmark of 406 tasks and 4,836 criteria. Training on its data noticeably improved…
Reflection releases Beam: 501B parameters and 30 points on the Intelligence Index
Reflection's first model is the second-best American open-weights model: 30 points on the Artificial Analysis Intelligence Index, using half the tokens of K2…
Conditional Program shrinks DeepSeek's n-gram memory from 203 to 51 GB at 2 bits
ERB: AI Agents Build Their Own Model Tests and Match the Reference Benchmarks in 75% of Pairs
MambaAct speeds up π0.5 robots: action expert 10x smaller, latency down 64.8%
LM2LM: Qwen3.8-27B Beats GPT-6-sol at Ideas for Improving LLMs and Is 33% Cheaper
Codex Teaches Team Sites to Update on a Schedule and Adapt to Each Visitor
OpenAI showed how to build Sites in Codex: scheduled tasks refresh the data, and users' own plugins give each visitor a personalized view of the page.
OpenAI Releases GPT-Live-1.1: New Voices in 6 Languages and Fewer "checking..." Pauses
GPT-Live-1.1 is now available to all users, with new voices for French, German, Hindi, Mexican Spanish, Korean and Japanese, plus more accurate accents.
Tadpole Proposes a Foundation Model for Physics: From Prediction to Understanding
Tadpole brings foundation-model logic to science: one model instead of separate networks for weather, airflow, material deformation and electromagnetism.
Artificial Analysis Launches AA-Robotics, Tests GPT, Claude and Gemini as a Robot's Brain
SpecTr-GBV at EMNLP 2026 Combines Multi-Draft and Block Verification to Speed Up LLMs
AIPOCH Open-Sources Open-Science, a Lab Notebook for "Vibe Research" with AI
Honeycomb Gives Video Models Constant-Size Memory for Loop Closure
New Survey Maps AI in Humanitarian Work: Six Tasks, From Weather to Refugees
A survey on Hugging Face Daily Papers describes AI as a tool for six humanitarian tasks: weather forecasting, energy, forests, poverty, post-disaster damage and…
ARC-SORT hits 81.82 HOTA on SportsMOT and runs at 50 fps
The training-free ARC-SORT tracker scores 81.82 HOTA on SportsMOT with the fewest ID switches among 69 entries, and runs at about 50 fps on a laptop RTX 3060.
GlobeBench: The Best LLM Correctly Predicts Environment Responses in Only 49.57% of Tasks
A Frozen 16 MB Vocabulary Matches a Trained One in Small LLMs
NVIDIA RTX Spark N1X Runs a Local Qwen3.8 27B Agent in 18.7 Minutes
Vera Rubin NVL72 beats GB300 by up to 5.7x in tokens per megawatt in AA-AgentPerf
Data Rewriting Beats DP-SGD on the Privacy-Quality Trade-Off
A controlled comparison on synthetic medical records with "canaries": fine-tuning leaks the most, DP-SGD hurts quality, and private rewriting offers the best…
Kolab: 115 Hours of Recordings Show That AI Agents Don't Understand Their Human Colleagues
Minnesota NLP researchers recorded 50 professionals from 40 occupations working with AI agents. Frontier models overestimate how well that collaboration goes.
DialFork measured procedural hallucinations in agents across 36,278 dialogues
The DialFork benchmark catches agents that sound smooth and factually correct but drift off script. It includes 36,278 dialogues, 24 tasks, 12 domains, and two new…
RealCLI: The Best Terminal AI Agent Scores 49% on a Test of Working With a Human
CoSTA Compresses Step-Audio 2 mini Speech Tokens with BPE: 10K–320K Added Units
OpenAI Engineer Hid a "Bug" in Toilet Paper and Got Caught Within 48 Hours
