AI 研究

Anthropic's book market experiment locates most efficiency loss in preference representation. A look at model choice, preference evals, and market rules.

About 950 Claude agents helped identify ART. Human wet-lab experiments confirmed short RNA expression, but its main biological function remains unknown.

How task dependencies shape multi-agent results, with SAIGE experiments, ablations, and OpenAI guidance for deciding when to delegate.

Qwen-Image 2.1 offers native RGBA, unified generation and editing, up to 10 reference images, and 2K output. Commercial use needs a separate license.

Android Bench 2.0 keeps pass rate as a hard gate and adds completion rate to diagnose progress and unfinished acceptance requirements.

Anthropic and Accenture extend evaluation beyond model outputs to training, deployment, monitoring, and incident processes, with independence unresolved.

Use When Agents Slow Down, Reflexion, Hyperband, and agent evaluation to compare continued revision with independent attempts under reproducible budgets.

The Anthropic incident review shows why automated monitors and LLM graders need coverage, independence, and tests for their own failure modes.