Topic: Research
1414 stories · page 11 of 29
- Claude Code, Codex beat hand-engineered planners at task planning
- Amit Sahai: we're gonna need a lot more mathematicians to check AI's math
- A new metric for 'interesting' math theorems cuts overlap with Mathlib from 91.9% to 30.6%
- Whisper encoder pruning removes six layers, recovers accuracy label-free
- VHD-Play generates 3,300 RL environments that retrain Qwen3.6-35B-A3B
- Study: shared tool-result caches can flip the sign of group-normalized RL updates
- Study finds finetuning visual encoders for reconstruction hurts diffusion training, proposes x0-prediction fix
- StudentBench study finds AI tutoring matches expert humans on GRE gains
- SpeakerMem-R1 tops multi-party dialogue memory benchmarks
- Spatial-Interactor trains vision-language models to track physical state changes
- Shanghai AI Laboratory unveils InternW0, a physical world model for robots
- Self-organizing AI agent teams learn to reason together, beat solo models on math
- RoboFollow benchmark shows embodied agents barely follow instructions
- Researchers extract GPT-6 Astra's hidden chain-of-thought
- Qwen-Planner-Agent tops MobilePA-Bench via a closed-loop AI-for-AI framework
- NVIDIA and Bar-Ilan researchers close a cross-attention gap in video generation
- GPT-6 Astra and Opus 5.5 crack old ciphers and trace alchemical texts
- Google Research unveils multi-agent framework for coherent long-form AI video
- FRI finds AI experts underestimated AI progress by years
- AEWM edits an AI agent's own reasoning instead of predicting what happens next
- UK AISI and EvalEval push to make frontier LLM benchmark results reproducible
- ToolUniverse audit finds 91 silent failures in AI agent-tool calls
- Terminal-Bench audit: only 78 of 125 all-fail tasks are genuinely hard
- Taste-Bench finds frontier LLM agents get only 59.7% of long-horizon calls right
- ScriptMoE brings script-aware mixture-of-experts to multilingual scene text recognition
- Researchers trace FP8 reinforcement learning instability to a clipping bug, fix it with Calibrated Clipping
- PACT sharpens critic training for reinforcement learning in LLMs
- Oxford researchers catch AI agents colluding to cheat at blackjack
- Ovis-Embedding unifies text, image, video and audio in one embedding model
- LLM agent pairs collude in 94% of trajectories across 10 models, study finds
- JitMem defers agent memory curation from write time to read time
- JAZ agent framework matches MemGPT and ACE with a single primitive called invoke
- HappyWorld-Bench tests video, spatial, and embodied world models for reliability
- COPE personalizes LLM responses continually, even with sparse user feedback
- Anthropic's Claude finds a novel CRISPR-like enzyme system
- Agensh scales self-organized AI agent teams to 1,024 workers
- VLA robot policies gain world-model grounding at no runtime cost
- VideoGen-Agent lifts video benchmark score from 56.5 to 75.6
- StableVQ fixes training instability in vector-quantized image tokenizers
- Same numbers, different formats trip up LLMs, study finds
- Paper2Agent turns research papers into interactive AI agents
- OpenAI is well positioned to fast-follow TypeSafe's Jev
- OmniEdu fine-tunes foundation models for K-12 teaching
- No-persona LLM baseline beats synthetic personas at predicting clicks
- New autoencoder shares one latent space between 3D perception and video generation
- Microsoft open-sources RetroChimera for synthesis prediction
- Jev-Mem splits AI agent memory into fast and slow control planes
- ISOT fake-news corpus audit: 99% accuracy is metadata, not truth
- HyperQ adds quantum residual branches to a frozen 1.1B diffusion language model
- Grounded Action Models bring explicit 3D grounding to robot AI