Daily AI news, sourcedUpdates / Research
Updates Research · 222 items · newest issue first · page 8 of 12
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- Reddit user says Nvidia's ICML spotlight paper DreamDojo has bugs in its released codeUnconfirmedr/MachineLearningOct 8Research
- Reddit user revisits the 'stochastic parrot' view of chatbots in 2026Unconfirmedr/singularityOct 8Research
- Reddit user predicts AI will make math 'liquid', discovering results in real timeUnconfirmedr/singularityOct 8Research
- Reddit user likens mathematicians such as Terry Tao to a scene from Good Will HuntingUnconfirmedr/singularityOct 8Research
- Paper proposes a faster diffusion model for estimating wireless MIMO channels under mobilityUnconfirmedarXiv cs.LGOct 8Research
- Paper's GraphOPD focuses distillation of LLM agents on the steps that actually enabled later progressUnconfirmedarXiv cs.LGOct 8Research
- Paper's KITE cuts communication between LLM agents by up to 36x by sharing one key layer's KV cacheUnconfirmedarXiv cs.LGOct 8Research
- Paper's GenST uses an LLM to forecast traffic at sensor locations with no past dataUnconfirmedarXiv cs.LGOct 8Research
- Paper proposes BRaVeS, a framework that reduces an AI agent's autonomy as its reasoning risk risesUnconfirmedarXiv cs.LGOct 8Research
- Paper tests five open-weight LLMs on harmful prompts disguised with emojis, leetspeak and encodingsUnconfirmedarXiv cs.CLOct 8Research
- Paper finds LLM sycophancy depends mostly on how hard a claim is to verify, and maximum reasoning removed it on hard puzzlesUnconfirmedarXiv cs.CLOct 8Research
- Paper uses LLM-generated explanations to detect fake news rewritten with different emotional tonesUnconfirmedarXiv cs.CLOct 8Research
- Paper's CoDR lets diffusion language models revisit tokens they committed too earlyUnconfirmedarXiv cs.CLOct 8Research
- Paper's Emo-Jev tests whether Jev's probabilistic judgments can handle emotion and sentiment classificationUnconfirmedarXiv cs.CLOct 8Research
- Paper proposes 'agent plasticity' to measure how well AI agents improve from experienceUnconfirmedarXiv cs.AIOct 8Research
- Paper's SAUCE estimates uncertainty for parallel multi-agent LLM reasoning without trainingUnconfirmedarXiv cs.AIOct 8Research
- Paper finds the same Agent Skills help some model setups and hurt others on 37% of SkillsBench tasksUnconfirmedarXiv cs.AIOct 8Research
- Paper's Route-Verify-Vote framework places second at SCoRE 2026 mixed-domain reasoning without fine-tuningUnconfirmedarXiv cs.AIOct 8Research
- NVIDIA researchers unveil PivotOPD to help AI agents recover from early mistakes
- Reddit user asks whether the argument that AI will hurt mathematics assumes humans must stay essentialUnconfirmedr/singularityOct 8Research