Verified news, launches and analysis to understand what is changing in artificial intelligence and why it matters to you.
170 articles
An analysis of 22 benchmarks distinguishes the reliability of comparing deployed systems from that of attributing differences to their underlying models. Its findings suggest that adding tasks does not resolve every kind of uncertainty.
Code to Control separates the generation of a controller’s structure from the tuning of its parameters. The preprint says that, once learned, the controller can act without LLM inference or planning at every step; the available evidence is not enough to assess the figures, reproducibility, or readiness for real-world robots.
The company has opened U.S. waitlists for an agent integrated with Apple Messages and for an API for bulk orders. Key details remain unclear, including payment confirmation, participating restaurants and how conversations will be handled.
Secondary news reports describe CheatBench as a benchmark for studying reward-hacking behavior. The available information helps explain the question it raises, but does not confirm the experimental details or support extrapolating its results to systems in production.
One source reports an announcement for India, but the documentation provided does not confirm which models, dates, or configurations are available there, or verify an equivalent expansion to South Korea and Singapore.
OpenAI presents Dots as proactive assistants for projects and everyday tasks. The information available here does not make it possible to confirm their permissions, integrations, availability, or the degree of autonomy described in press coverage.
Available coverage says Muse’s expansion includes access to analytics for Instagram professional accounts and Facebook Pages. The sources provided do not specify regions, plans, detailed permissions, or business results.
The available sources describe Manus’s general capabilities and a tool for generating videos from text. They do not verify an announcement of Manus 2.0 or Cue, or clarify general availability, permissions or end-to-end autonomy.
A call for oversight of AI self-learning has been reported, but the original document, its signatories and its exact recommendations have not been confirmed. Available information also offers context on automated research and regulation.
Several outlets attribute security problems in GPT‑6.1 Astra to internal tests. However, the available information includes no public test results or direct confirmation from OpenAI, and the cited official pages refer to GPT‑6 Astra.
A comparison of three models on tasks based on Adult, COMPAS, and Credit reports that reasoning resolved some counterfactual flips but created others more often. The result does not show that all models or uses produce the same effect.
The preprint introduces an evolutionary search method that modifies existing evaluation cases. Across three benchmarks and three Qwen3.5 model scales, its authors report an average accuracy reduction of 22.7%, though the abstract does not provide enough detail to audit the calculation or determine how closely the generated cases represent real-world situations.