Search Results for "surface"

...Instead of building bespoke evals, Inspect Petri automatically generates audit environments from seed “special instructions,” orchestrates an auditor model to probe a target model, and simulates tool use and rollbacks to surface risky behaviors. Each interaction transcript is then scored by a judge model using a consistent rubric so results are comparable across runs and models. The system supports major model APIs and comes with starter seeds and judge dimensions, enabling minutes-to-insight workflows for questions like reward hacking, self-preservation, or eval awareness. ...

...The project includes validation stages such as scorecards, dogfooding, proof-of-behavior checks, and optional live read-only smoke tests. Overall, CLI Printing Press aims to turn any useful web service into a practical automation surface for both humans and AI agents.

...It breaks down complex concepts such as planning, tool usage, memory management, and multi-step reasoning into digestible explanations and practical examples. The project is organized as a learning resource rather than a standalone framework, making it particularly useful for developers who want to move beyond surface-level prompt engineering into full agent system design. It explores how agents interact with environments, execute tasks, and maintain context over time, highlighting both strengths and limitations of current approaches. The repository likely includes diagrams, annotated code samples, and conceptual walkthroughs that mirror real production systems.

...It also supports a “Chat-Supervisor” pattern where a lightweight realtime chat agent handles user interactions and delegates more complex reasoning or tool usage to a stronger textual model (e.g. GPT-4). Because realtime agents are still a beta feature, the code and API surface are subject to changes and may evolve.

...The idea is that AGENTS.md acts as a “README for agents”: a predictable, structured place where humans can put instructions, conventions, build/test commands, environment setup, and other guidance that generative agents (e.g. code-writing, code-assisting tools) should consult when operating in the repo. Instead of putting everything in README or doc files (which are more human-oriented and might mix high-level narrative), AGENTS.md is intended to surface agent-relevant details that help them “do the right thing” (tests, style, project structure, tooling).

Bubble Pond is a simple artificial life program that simulates bubble organisms that bounce around on the surface of a pond, eating and digesting different types of food, and reproducing both sexually and asexually.

In this project we develop algorithms that can solve a cooperative problem between four agent types, and their goal is to repair a surface time-efficiently, and a multi-agent simulator application to parametrize, animate, and test the solution.