Topic: Research
1414 stories · page 7 of 29
- Δ-MOPD distills several teachers by transferring logit shifts, not endpoint policies
- MIT Technology Review: humanoid robot hype outruns what chatbot-style AI can deliver
- MIMESIS builds a 9B user simulator for training interactive agents
- Mathematicians' association urges OpenAI boycott after release of AI-written proofs
- LoopOPD lets looped language models learn from their own deeper loops
- Learn2Play Bench tests whether LLM agents learn from experience in new text games
- Iris-3B pixel-space model rivals latent models, but fine-tunes show no gain
- Google Research workshop report maps open problems in agentic AI privacy and security
- DreamTrue robot world model takes first place in AgiBot World Challenge 2026
- AgentGarten pairs game engines with a neural renderer to train agents
- Zuckerberg-backed Biohub coordinates $1.8 billion push for AI models of cell behavior
- TRIAGE stabilizes native NVFP4 reinforcement learning for LLMs
- SWE-Game benchmark: coding agents score under 60/100 on building Godot games
- SGF+ gives autoregressive video models separate parameters for context and denoising
- Recurrent Looped Transformer generalizes parity to 256 bits
- R-Quest keeps self-evolving reasoning models from collapsing by filtering bad questions
- Paper says Lean version of OpenAI's Navier-Stokes proof does not match the original
- Paper proves matching bounds on state needed for long-memory forecasting
- NVIDIA says fine-tuned Nemotron 3 hit gold level at IMO 2026 and IOI 2026
- Long-WAM lifts robot success with 19.2 seconds of video context
- KVFetch adds position-based prefetching to KV cache compression
- KLPO proposes critic-free RL for LLM agents with one rollout per prompt
- HiPHI white paper presents 617.5-hour motion capture dataset for humanoid robots
- GRACE compresses video autoencoders, cutting Wan2.1-I2V-14B tokens 8x
- Google Research trial: AI sharpens senior lawyers' judgment, not juniors'
- WING moves human egocentric video skills to robot policies
- TRACE brings FP4 rollouts to MoE reinforcement learning, up to 5.4x faster
- Refusal steering in four open-weight LLMs traces to sparse components
- Paper splits causal influence in language models into capacity, responsiveness and alignment
- OpenAI releases new math results from an internal model, with Lean proofs
- OpenAI posts preprint titled "Integer multiplication below n log n"
- Microsoft Research Podcast: Jennifer Neville on testing AI beyond benchmarks
- LMBuild benchmark tests whether LLM agents can design buildable structures
- LLM-as-Jev paper: Qwen3.5-4B already works as a decision model without training
- Latent protein languages PLL and SLL help transformers scale in protein generation
- Jev-style typed decision models follow labels over definitions, paper finds
- Google Earth AI's PDFM location embeddings tested on five public health tasks
- Empirical Variational Autoencoder (EVA) learns its own latent priors for sequence generation
- DeskForge builds a 1.2M-observation desktop dataset to train computer-use agents
- AutoSciBench generates and revises scientific-agent benchmarks automatically
- Activation alignment lets tabular models TabPFN-3 and TabFM close much of the gap with less context
- VeriHarness turns an LLM into its own verifier for long-horizon agent tasks
- Swapping the gold reference moves a six-language PII benchmark score by up to 7.55 F1 points
- Spatial Memory Intelligence uses an MLLM to manage memory in long-video world models
- Researchers argue human-in-the-loop checks on AI agents push humans out
- RealCompanion benchmark finds AI companions rarely need past messages
- Queen, a 4B chess language model, explains its moves at Grandmaster level
- OpenAI's Navier-Stokes claim splits mathematicians at Heidelberg forum
- NAVA-WAM pretrains a robot action policy on videos without action labels
- Multilingual GSM-Symbolic maps what drives cross-language transfer