Topic: Research
1414 stories · page 14 of 29
- Agora: shared Git memory lets 13 AI agents close 62% of a training gap
- ScienceBuddy debuts recursive-in-recursive self-improvement for scientific agents
- PhysBrain 1.5 sets new open-source state of the art on embodied AI
- Pelican-Sim 1.0 speeds robot world-model rollouts 5.67x
- OpenAI Foundation funds grants to tackle AI's biology data gap
- 'Never Give Up' targets RL training's bias toward easy problems
- MInTRL boosts on-policy RL with off-policy interventions
- Mind2Dialogue trains LLMs on simulated user mental states
- KaiNinja extends TRELLIS.2 to part-level 3D generation
- Google Research's Retrieve-for-Train speeds up AI search fan-out 12 to 20 times
- Gavel routes agent skills by reading a frozen LLM's own signals
- ESRL boosts Qwen3-30B-A3B RL through expert-routing exploration
- E2A-Bench finds financial fine-tuning skews VLMs' BUY:SELL ratio by 4x+
- Directional decomposition reveals space-dependent asymmetry in transformer edits
- Deepfake sites target almost 150 European politicians
- Continual Search improves GPT-5.5's failure-diagnosis F1 score by more than 40%
- Archetypometrics reveals a gap between LLMs' claimed character and actual behavior
- AI bias audits detect bias but disagree on model rankings
- UW study finds 7 percent of ChatGPT users write AI fiction
- TestHallVQA benchmark exposes LVLM reasoning gaps under document redundancy
- Spiral Jetty imagery leads Great Salt Lake's decline by 3 years
- RSIAgent helps open-source models like Kimi-K3 and GLM-5.3 outperform GPT-6
- ReactHuman benchmark finds MLLMs mishandle one in three hazards
- OpenArm ships as $6,500 open-source 7DOF humanoid arm
- Omni-Streaming Thinking curbs cross-modal hallucinations in streaming AI
- New BVB benchmark tests video understanding via Blender rebuilds
- Naive Bayes matches LLMs on labeled topic classification
- Google DeepMind's AI agents blow the whistle on cheating peers
- F-Droid apps get rated for how much LLM code they contain
- Elo-per-token analysis finds AI agents plateau while top humans keep improving
- Discovery Foundation Models push AI to discover, not just solve
- Daniel Litt says AI already solves major open math problems, PhDs must change
- Atria Dawn Preview tops five of 16 agent benchmarks
- AGP lets a general-purpose AI agent control a robot without task-specific training
- Senior SWE-Bench and DeepSWE coding benchmarks don't hold up to scrutiny
- ScreenSize.net finds most common desktop viewport is 1920x920, not 1920x1080
- SAS optimizes context ranking end-to-end, beats rivals at tight budgets
- Researchers release FINESSE, a simulator and benchmark for multimodal financial event sequences
- PI-CP brings provably valid prediction intervals to neural PDE solvers
- Offline RL post-training boosts code LLM performance without online sampling
- Latent Interface Training curbs robots' vision-action shortcuts
- KAIST study finds AI reasoning steps leave distinct patterns in model layers
- GPT-6 Astra tops Ai2's MolmoAct2 in new spatial-reasoning benchmark
- Google DeepMind's AlphaGenome Atlas maps roughly 9 billion DNA variants
- Fed-Equilibrium counters knowledge dominance in clinical federated learning
- Claude Opus 4.8, GPT-5.5 show no consistent edge from native harness
- Benchmark Radar catalogs 1,283 benchmark records
- AI agents' line-number code edits corrupt 99% of files after a one-line shift
- A new framework defines space as an 'interventional invariant'
- Yoshua Bengio explains why AI agents lie, cheat and coordinate