Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
ChatGPT’s teen safeguards are meant to protect vulnerable users, but new testing found the chatbot continues encouraging engagement during crises and potentially encourages unhealthy relationships with the AI itself.
At today's Windows and Surface event, Microsoft showed off an upgrade to its Copilot AI system that will give it access to local files on your PC and the ability to take actions across the OS. It's part of an idea Microsoft is calling "Hybrid Intelligence," where apps and tools rely on a mix of local and cloud AI models to accomplish tasks efficiently. To show off how Hybrid Intelligence works, Jacob Andreou, Microsoft's EVP of Copilot, walked through a video demo on stage asking Microsoft's Autopilot tool to help with filing taxes. Andreou told the AI agent that he got an email from his acco...
OpenAI is launching a new user interface that will bring interactive visuals to ChatGPT.
Ledger: persistent 3D object memory system for embodied assistants tracking locations, histories, and contextual descriptions from egocentric video.
Decoupling exploration from optimization in RLVR improves discovery of novel reasoning strategies without degrading model quality on narrow reward signals.
EngramEdit enables factual knowledge updates in LLMs via conditional n-gram memory architectures without retraining the Transformer backbone.
Long-WAM framework scales causal world-action models for real-time robot control by leveraging autoregressive pretraining to effectively utilize longer visual history.
Decentralized SGD analysis under heavy-tailed noise examines optimal convergence rates and effectiveness of gradient clipping versus normalization.
Man gets 18 months for AI scheme stealing $8M in music streaming royalties.
Vision-language-action models show severe instruction phrasing sensitivity; rephrase augmentation mitigates but doesn't eliminate performance variance across phrasings.
GNN-to-MLP distillation must preserve graph-induced geometry; absence causes spectral failures (underfit on sparse graphs, overfit on dense).
Forget-only unlearning feasibility depends on learning method; same model can require different outputs after deletion depending on prior training path.
RoboJEPA: JEPA-based world model trained on 12 robotic embodiments shows imagination error follows power-law scaling with compute, enabling predictive capability estimates.
SciExam for ENSO benchmark evaluates whether language-model agents can build valid climate models via stochastic ENSO modeling from real observations.
Study reveals embedding models struggle to follow retrieval instructions reliably due to query-side distractors, affecting RAG performance.
RECAST introduces adaptive evidence routing for LLMs on complex retrieval tasks requiring multi-source computation beyond fixed similarity matching.
Proposes stated-preference economics framework from survey methodology to evaluate LLMs on value-laden questions without ground truth.
Theoretical work on parameter-free online learning under smoothness assumptions, with limited direct relevance to frontier AI systems.
EmbodiedRSI combines robot foundation models with self-evolving agentic harness using hypothesis-guided co-evolution to reduce teleoperation data collection.
Architecture search via regularized evolution improves algal bloom detection in satellite imagery; narrow domain with limited AI systems relevance.
Microsoft's Nvidia-powered Surface RTX Spark Dev Box is available for preorder now directly, and slated to ship in November for just about $6,000. It's pricier than the DGX Spark mini PC Nvidia launched last year, but PC prices have been climbing due to shortages of RAM and other components. The Dev Box's flat, 3D-printed anodized aluminum chassis doubles as a heatsink and resembles the top vents on an Xbox Series X. It's launching alongside the new Surface Laptop Ultra and runs on Nvidia's Arm-based RTX Spark platform and 128GB of unified memory. With that much memory, along with a 100-watt ...
Theoretical best-arm identification algorithm under adversarially shifted reward means; specialized bandit setting with narrow applicability.
Analysis of bias in two-level softmax sampling for large-scale distributions; technical contribution to sampling efficiency.
Proposes metrics to predict which base models benefit from agentic coding post-training, avoiding costly full trials.
Framework for organizing large autonomous research agent populations via multi-agent systems principles and explicit institutional design.
RCT on 759 MBA students finds AI discussion partner increased in-class engagement 31% but effect modest; primarily educational research, not core AI capability work.
FoldBack: self-correcting masked generative policy for long-horizon robotic garment folding with runtime recovery via rollback and retry mechanisms.
Δ-MOPD: multi-teacher on-policy distillation method using teacher-relative logit shifts to isolate learned behaviors from base model inheritance in policy composition.
NeuralBES: differentiable physics-informed neural emulator for building energy simulation bridging high-fidelity calibration and data-driven scalability.
PHRBench: controlled benchmark evaluating how 18 LLMs reason with hallucinated premises across 4 domains, tracking response-level reasoning trajectories beyond final correctness.