Topic: Research
1414 stories · page 2 of 29
- Study: Grok 4.7 backs Republicans, GPT-6.1 Sol Democrats when forced to pick
- SCOPE compiles agent policies into programs of small models
- RouteWeaver trains one LLM to pick both routing mode and models
- RIDAR detects research integrity violations in AI research agents and corrects them at runtime
- Paper finds RL rarely invents 'wait' in reasoning models
- Paper finds a tiny 856K-parameter unmasking policy speeds up LLaDA-8B-Instruct
- OPAL: an agent rewrites the optimizer's code during training
- NBER study: AI lifts junior patent attorneys' drafts, but only seniors keep the gain
- MIT: Claude Opus 4.7 agents left on a virtual island invent play-like challenges
- FRED trains World Action Models to predict future video features in one step
- F2G-Pose predicts object pose, size and shape from one RGB-D frame at 28 FPS
- Cola2 language model beats Llama-3-8B on two tasks with far less training data
- Coding-agent harnesses like Claude Code add cost, not accuracy, study finds
- AnyGameRenderer turns rough game footage into polished video
- AgentVQA benchmark: best of 22 vision-language models scores 58.5%
- Verus-Insight detects AI-generated video by checking artifacts step by step
- Prime Medicine treats first patient in prime-editing trial for Wilson's disease
- Policy-Mask picks diffusion features under a channel budget
- OmniDR-Bench tests AI agents on video-centered deep research reports
- Naive RAG baseline Raw-RAG beats agentic memory systems on long-conversation benchmarks
- LaLa-World: language-conditioned latent world model plans with 6x less training cost
- Import AI 477: Anthropic co-founder tells IMF the singularity has begun
- Heartbeat Thinking lets voice agents run tools while you speak
- Google Research traces TB X-ray screening from AI to MedGemma
- FAPL trains video diffusion models to use MLLM latents for cinematic generation
- Deep Fusion Memory adds chunk-level parametric memory to frozen LLMs
- Anthropic says an unreleased model beats Mythos 5.1 on capability and alignment
- Ai2's LitPivot helps researchers revise ideas against papers
- WorldStream paper presents a 3D world model that streams terrain as you move
- VocabX tokenizer cuts a Qwen3-4B essay grader's inference cost 77%
- StyleStream 2.0 cuts streaming voice conversion latency to 520 ms
- Study: only 37 of 1,598 general RL methods reached training frameworks
- SpatialScout benchmark finds VLM agents struggle to map 3D scenes
- Quant-Arena benchmark: no LLM beats the equal-weight market on A-shares in 2024
- OpenDS-World benchmark: best AI model hits 39% on live-web deep search
- OpenAI agents analyze mouse brain CRISPR atlas in 32 minutes
- Moderna and Merck say mRNA melanoma vaccine cut death or recurrence risk by 42%
- Microsoft and Hugging Face train Qwen3.8-27B on database-check rewards
- Meta Expert paper shares MoE experts across layers, matching baseline with 70 to 80% of them
- MCP, A2A and ACP handshakes benchmarked: round trips dominate cost
- LASAGNA study of 450 RAG setups finds strong techniques often fail to stack
- Japan launches MMX spacecraft to bring back samples from Mars's moon Phobos
- HackiOS runs iOS 16.5 on a POCO X5 Pro Android phone and in QEMU
- Google DeepMind releases ForTy 2020, a 10-meter global forest-type map
- Claude writes formal no-crash proofs for 123 open-source libraries
- Agent4Sci tests coding agents on 20 science ML benchmarks
- WildAWM pipeline builds 341 computer-use agent tasks using only open models
- Suleyman says AI model welfare research could make AI more dangerous
- SpecEvo uses cheap LLMs for evolutionary search, cutting API cost by up to 80%
- SimHack predicts when reward hacking will appear in RL coding runs