Releases and updates linked to profiles, context and primary sources.
The five-year budget request proposes combining machine-learning algorithms with contactless sensors. The official document describes development and validation goals, not evidence that the system can reliably detect lies.
A theoretical and computational study suggests that an agent can influence what a debate overseer learns by selecting and ordering true arguments. The result is a proof of concept, not a demonstration involving deployed models or human overseers.
Microsoft has announced plans to bring several Copilot experiences together in one app for consumers and businesses. The proposal combines chat, coding and agents, but important questions remain about availability, permissions, integration and price.
A study of multi-turn agents argues that a student can learn to act as though it knows information it never received. Its authors propose Privileged Self-Practice, which keeps that information in the practice context rather than incorporating it into the training loss.
A preprint compares global skill retrieval with retrieval limited to the family in which each skill was validated. In its tests, the second approach achieves higher hidden-trajectory utility; further details are needed to interpret the metric and generalize the result.
The preprint brings together resolved events and dated news to evaluate agents at different points in history. Its abstract reports that research tools improve scores across 12 models, although every model still trails historical market forecasts.
The benchmark evaluates agents that attempt to reconstruct results from machine learning papers using released materials and a fixed GPU budget. Its success rates show increasing difficulty, but do not determine whether the papers’ scientific conclusions are valid.
A preprint describes a system that retrieves domain-specific definitions of sensitive information at inference time. Its initial results favor its detection recall over a reference classifier in most tested combinations, although the information available does not make it possible to assess all error types or its readiness for real-world use.
The benchmark brings together 799 Indian retail banking cases and examines safety, tool use, responses, and advice. Its authors argue that getting a case right once is not enough to measure reliability.
Akamai confirmed a multiyear commitment to provide computing capacity for Anthropic, with a focus on CPU workloads. Earlier reports initially cited $1.8 billion; a later announcement put the expanded commitment at $11.6 billion. The terms do not, by themselves, show that all of this capacity is already operational.
AI benchmarks have moved from scoring answers in datasets to testing systems on tasks and in interactive environments. This shift broadens the evidence available, but it does not turn a score into a complete measure of usefulness or reliability.
A price per million tokens is not enough to budget a multi-step workflow. Learn how to add up input, output, cache, retries, and review, and set limits before you run it.
News that modifies the map of models and capabilities.
The launch requires reviewing evaluation, deployment and security.
Border model ↗Code and knowledge tasks with effort checks.
Knowledge ↗Multimodality, code and agents with a low latency profile.
Stable version ↗A news story does not end in the headline.
Relates the ad to version, access and modalities.
Identity ↗Distinguish stable, introductory rates and missing data.
Cost ↗Review conditions and comparability before accepting a benchmark.
Evidence ↗