Topic: Research
1414 stories · page 10 of 29
- QwenGyre framework speeds up RL training for xlong-horizon LLM agents
- Paper predicts encoder-free multimodal LLMs catch up at about 10^22 FLOPs
- OMP-MoE prunes MoE LLM experts without training, keeping 93.3%
- Nicholas Polson posts 258 SSRN papers in 2026, apparently AI-assisted
- NASA tests Claude and generative AI to give spacecraft more autonomy
- Multiverse Computing's ProvenanceGuard checks the source behind MCP agent claims
- Microsoft Research introduces Quine, an AI system for biology
- MaD-RL uses RL to make LLM outputs match a target distribution
- LSPD brings RL-style exploration to on-policy distillation of LLMs
- Imprint Reader describes model weight updates in natural language
- Google Research's Diffusion Controller steers image models with a small add-on network
- GAGAR paper: agentic grader ranks passing code, reweights RL rewards
- DN-MOPD rescales teacher feedback to improve multi-teacher distillation on Qwen3.5
- DISCO splits long context across worker LLMs to curb context rot
- Continual learning with no offline phase: local replay reaches 91.6% on split-MNIST
- ColNanoVDR shrinks multi-vector document retrieval query encoders without using any pages
- Bearing fault study: input representation, not alignment method, decides domain adaptation
- Associative-algebra layer speeds up a 110M Transformer by 6.2 to 7.8% but lowers scores
- Anthropic says Claude found a Crispr-like system; scientists urge caution
- AnswerMap shows where a VLM looks using only its yes/no answers
- Allspark paper: a weak model's reasoning steers stronger models
- AI Night-Scientist uses GRPO to train LLMs toward more creative research ideas
- WhatWorkedBench tests whether AI research agents actually understand their own experiments
- ViRDM drops the teacher-critic stack for few-step video diffusion
- Sparse autoencoders encode parts of speech as distributed latent groups, study finds
- OpenAI's GPT-6 Astra hits 80% on IKEA assembly error benchmark
- Nvidia's SoL-Pi system cuts coding agent token usage nearly in half by optimizing the harness
- MemBodied gives robot vision-language-action models a fixed-size memory
- Jev detects AI alignment failures zero-shot, 63x cheaper than LLM judges
- IDQD-RDO lets video codecs optimize directly for perceptual quality metrics like MS-SSIM and LPIPS
- DeltaWAM trims compute for bimanual robot world-action models
- Calibration as a first-class criterion in LLM evaluation
- AV-GRPO brings reinforcement learning to joint audio-video generation
- AI access makes people almost entirely unwilling to say "I don't know," study finds
- A paper argues the linear representation hypothesis needs a group action
- A new survey maps how memory works in autoregressive video generation
- WROP dataset tests object permanence in 14 video world models
- World Action Agent lets VLMs pilot robots, hits 75.6% on LIBERO-Pro
- Uranus: a diffusion-based robot simulator that streams video at 24 FPS
- Study finds transformer LLMs can output a superposition of two predictions at once
- Rust gets a 'Golden Spike' to share generics with Evan Ovadia's Valen
- Rufus-Air recreates GLM-4.5-Air's post-training in eight open stages
- RGBD20K: a 20,000-pair RGB-D segmentation benchmark with 160 categories
- OmniEcho brings spatial audio understanding to embodied agents
- MemoryAthena routes language models between retrieval and generated memory
- IterSynth: a role-decoupled search agent design that beats prior 8B agents by 4.2%
- GeoPair compresses transformers without retraining by pairing layers geometrically
- FLEET adds memory to LLM sampling, cutting compute 3x
- ExplorationBench tests 10 AI systems on rule discovery in verifiable 'Alien Worlds'
- EmbodiedSWE turns coding-agent solutions into training data for robot policies