Daily AI news, sourcedUpdates / Research
Updates Research · Unconfirmed · 138 items · newest issue first · page 1 of 7
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- WarcraftSim, a self-play agent trained on one desktop, beats Warcraft III's built-in insane AI in 89% of gamesUnconfirmedr/MachineLearningOct 19Research
- TempoPFN, a linear-RNN time series forecasting model pre-trained only on synthetic data, is accepted at NeurIPS 2026Unconfirmedr/MachineLearningOct 19Research
- Researcher traces where LLM "slop" writing style lives in Qwen3-4B-Instruct and which training stage installs itUnconfirmedr/MachineLearningOct 15Research
- Developer's sparse-autoencoder tool finds evaluation awareness and mild environment exploitation learned during MiMo-v2.6 RL trainingUnconfirmedr/MachineLearningOct 15Research
- Event2Vec library learns additive event embeddings so a linear score on a history splits exactly into per-event termsUnconfirmedr/MachineLearningOct 14Research
- VH-GPO restores learning signal in GRPO agent training when every rollout in a group failsUnconfirmedarXiv cs.LGOct 14Research
- Paper splices hints into reasoning traces offline so models learn from a demonstrator's failed proofsUnconfirmedarXiv cs.LGOct 14Research
- Predict Before You Modify has self-evolving agents forecast the effect of changes before costly evaluationsUnconfirmedarXiv cs.LGOct 14Research
- LACRM uses LLMs to mine directed competition relationships between advertisers in e-commerce adsUnconfirmedarXiv cs.LGOct 14Research
- MDN-VJEPA represents multiple possible futures in JEPA world models with a mixture density predictorUnconfirmedarXiv cs.LGOct 14Research
- Study tests whether agents can improve LLM data-engineering strategies from experimental feedbackUnconfirmedarXiv cs.CLOct 14Research
- Fin_Agent pairs rules with zero-shot Qwen2.5-32B for multilingual financial NLP at FinEval 2026UnconfirmedarXiv cs.CLOct 14Research
- ChronoRegime tests whether LLMs know when the future is predictable enough to commit to an answerUnconfirmedarXiv cs.CLOct 14Research
- Loose Lips benchmark studies how LLM agents leak private information in agent-to-agent dialogueUnconfirmedarXiv cs.CLOct 14Research
- PsyBer-Agent simulates recommender-system users with fatigue, emotion and shifting preferencesUnconfirmedarXiv cs.CLOct 14Research
- Long-Horizon-Terminal-Bench 2.0 expands to 77 terminal tasks across 13 domains to measure sustained agent progressUnconfirmedarXiv cs.AIOct 14Research
- Correction Training targets exposure bias in diffusion modelsUnconfirmedarXiv cs.AIOct 14Research
- ClawLoop trains harness-based agents through a self-improving loop inside their own sandbox environmentUnconfirmedarXiv cs.AIOct 14Research
- The Silent Failures in Agentic Data Science looks past accuracy and code correctness when judging data science agentsUnconfirmedarXiv cs.AIOct 14Research
- Artificiality's First Exam proposes an LLM-referenced C-test for evaluating AI beyond human levelUnconfirmedarXiv cs.AIOct 14Research