Topic: Research
1414 stories · page 6 of 29
- Sakana AI's MASS trains Qwen3.6-27B on its own best team workflows
- RSI-Anything uses predictive world models to steer recursive self-improvement
- Representation Guidance pushes pixel diffusion transformers to FID 1.55 on ImageNet
- ProgramGym offers 1,648 training environments for coding agents
- On-policy distillation paper: training-induced differences transfer, scale-only ones barely do
- MIRA agent refines text-to-music prompts against rubrics of user intent
- LatentGRM judges in latent space, cutting inference time 6.1 to 7.0x
- InstantFusion links denoising across frozen image generators
- Hebero benchmark trains one robot policy on 40 manipulation tasks
- Epoch AI finds GPT-5.6 Sol and Claude Fable 5 overstate research results
- Causal Market Simulator estimates market impact as a causal effect with a prior-data fitted network
- Algebraic integer quantization (AIQ) is competitive with MX for LLM inference, paper claims
- The Lattice of Transition Laws: diffusion and AR decoding as paths on one lattice
- SPW-Nav world model streams 2K 360-degree video from language instructions
- SpaceFlow generates 3D objects with per-part control levels, no training
- Skill Constellations maps how agent skills are copied across GitHub
- Opera, a verbal critic for coding agents, lifts resolve rates by up to 15 points
- On-policy distillation teaches reasoning skill, not new facts, paper finds
- OmniCapBench tests audio-visual captioning across 786 videos
- ME-World generates synchronized first-person video for multiple interacting agents
- Incremental-OEDR updates deep research reports instead of rewriting them
- GPD distills 3D geometry cues into RGB-only vision-language models for spatial reasoning
- FreeMatching finds dense correspondences across image edits where motion priors fail
- Foundations of Large Language Models: a book on six core LLM topics
- CARE certifies VLA inference speedups with failure-risk guarantees
- Byteification retrofits subword LLMs into byte-level models
- AgenticBBO-Bench tests LLM agents on black-box optimization in five domains
- WorldGuide, a closed-loop video world model, reports higher task success than MiniMax-H3
- U-Space maps token-level uncertainty in language models without training
- Thomas Hales on Lean: AI autoformalization is practical, soundness bugs fixed
- Station agent environment rediscovers 62.7% of findings from three ICLR papers
- SpecFold speeds up diffusion LLM speculative decoding by up to 1.99x
- reViT paper claims one recurrent block matches a full-depth vision transformer
- REMORY adds soft memory tokens to LLM context compaction
- Paper: higher LLM agent accuracy does not mean more epistemic humility
- OuroWorld turns static 3D Gaussian Splatting scenes into looping 3D cinemagraphs
- OneSearch-VL unifies image and video deep research in one 8B agent
- MiMo-V2.6 report details scaling RL compute along three dimensions
- Memento 3 paper: frozen-LLM agent clears all 25 public ARC-AGI-3 games
- MC-Sparse paper reports 1.8x to 2.3x faster diffusion transformers, no training needed
- Language models get derailed by changes users later rejected, paper finds
- Framework lets frozen medical LLMs and VLMs learn from deployment, with gains up to 34.2%
- Code-Only-as-Policy (COAP) paper drives robots with code, no VLM or VLA at test time
- Anthropic's Claude Science maps the whole sky in ultraviolet for the first time
- Trace2Env lets an LLM agent act as the environment, built from historical traces
- Synthesis Through Simulation: an LLM agent generates enterprise data via policy-enforcing APIs
- SQAM fine-tunes flow policies with a scalar adjoint, skipping per-step Jacobians
- SpaceCast-Bench: best of 21 vision-language models scores 58.0% on predictive spatial reasoning vs 87.2% for humans
- Self-Retrospection Distillation lifts RLVR agents by up to 24.2 pp
- SanSi turns a looped language model into a decision model with 72.0% accuracy