Topic: Research
1414 stories · page 5 of 29
- Study finds best OpenAI models now solve about 35% of theory problems, up from 2%
- Simplified Matching Distillation targets 4-NFE diffusion with regression only
- Player-1 trains computer-use game agents with RL on 16 synthetic environments
- Paper finds most wrong actions by fast agent models follow a correct reading of the state
- LOTUS pretrains time-series language models to forecast the next patch
- LatentLoop looped MoE predicts 19.0% compute savings over vanilla MoE
- Inverse reasoning model lets Qwen3-0.6B self-distill from answer-only data
- HybridAgents paper mixes latent reasoning with text tool use, claims 10.4% gain
- Google DeepMind sees no AI-driven speedup in AI research, engineer says
- CriticOPD has teachers critique student errors to fix on-policy distillation
- ATAS gives agentic video generation scene and event memory, tested on four generators
- Ai2 releases HiRO-ACEv2 for 3 km extreme weather at about 1/300th the cost of X-SHiELD
- ACCRUE continual off-policy RL cuts real-robot training time by 31.0% to 68.5%
- Yau Laboratory Alliance launches in Beijing to push AI into math research
- Vela turns real user behavior logs into LLM training tasks
- UniForge3D joins 3D reconstruction and generation in one shared space
- UltraBERT reaches 91.0 on GLUE and 47.7 on BEIR with a new encoder training recipe
- Tracking-Value PPO (TPPO) reports 7.0 and 8.4 point math gains over PPO on Qwen3
- TerminalHorizon engine synthesizes long-horizon terminal tasks, tunes Qwen3.5-27B
- Paper: uncertainty-guided contraction cuts reasoning SFT budget to as little as 0.22%
- Minions pools idle Apple Silicon GPUs into a distributed inference cloud
- LeWAM trains one JEPA for both robot policy and planning
- Import AI 476: AI visual reasoning trails humans, and new open models trail China's
- Harness^2 lets LLM agents recursively improve their own harnesses without verifiers
- DeltaTTG trains multi-turn agents with a Token-to-Go progress meter, beating GRPO
- ClawLoop trains harness-based LLM agents to self-improve without a teacher model
- BranchOPD adds branching to on-policy distillation and beats Vanilla OPD
- AgentSWE benchmark finds no coding agent delivers the agent a requirement asks for
- Agent Sociodrama tracks social dynamics in LLM agent teams and recasts personas to fix failures
- TouchScale pairs 500 hours of human video with touch data
- SynthSelf lets a slightly larger model rewrite data to post-train small models continually
- Study of 127,833 clinician queries finds LLM benchmarks miss real use
- StudentSim trains AI student simulators to test AI tutors before real learners
- Sponsored white paper: humanoid robots learn pro tennis style with split planner and tracker
- SoftwareGym synthesizes 182 GUI apps to train computer-use agents
- Semi-OPD beats on-policy distillation in 14 of 17 model pairs
- ReHarness adds backward revision to LM agent memory for continual learning
- PreSTO prefetching speeds up LLM power sampling by a median 1.7x to 2.3x
- Paper shows encrypted backdoors can hide AI capabilities from elicitation audits
- Paper finds S-Space, a linear spatial workspace inside multimodal LLMs
- OpenAI's claimed faster Fourier transform is impractical, a blog post argues
- Microsoft Research's Fork-and-Flush wipes chat context to unstick AI agents
- Elastic Latent Recurrence removes the one-loop tax in looped language models
- DiffusionART reports about 12x fewer GPU hours for RL on diffusion models
- DataVista benchmark tests 19 multimodal LLMs on data videos; best scores 70.0%
- Data-space iteration removes timestep discretization from few-step generation
- AAArena tests whether AI agents can learn to top human game ladders
- VisionWeave saves 43% of visual tokens on average in Qwen3.8-27B
- SWE-Duel arena has coding agents plant and fix each other's bugs
- SepGen generates video with one audio stem per sound source