Search
Find blog posts, paper notes, and projects
Separate keywords with spaces. All keywords must match; Chinese and English aliases such as RAG and 檢索增強 are supported. Title matches appear first.
- How Did an Agent Evaluation Reach a Live Website? Consent and Egress Boundaries in the Epoch Incident
Two first-party reports from Epoch AI and Trajectory Labs show how evaluation agents reached a live website; this article separates task design, network isolation, and monitoring.
- MKA: How Multimodal Agents Complete Professional Work
Google DeepMind's MKA evaluates multimodal knowledge agents on end-to-end tasks spanning documents, charts, and spreadsheets. This reading explains Climb scoring, the vision ablation, and limits on reproducibility.
- Gemini Decisions API: Routing AI Agents with Probabilities and Confidence
Explore Gemini Decision 1's three question types, a Python routing example, confidence thresholds, and multimodal limits.
- Jev Beat Pokémon Red: Was It the Model or the Harness?
Using Jev's Pokémon Red run and public code, this article separates the decision model's role from game state, pathfinding, and recovery mechanisms in a long-running agent.
- How to Use pstack: poteto-mode, Workflow Basics, and Six Practical Examples
A source-grounded pstack guide covering poteto-mode, model setup, and six prompts for debugging, features, refactoring, investigation, review, and design, with Cursor, Claude Code, and Codex boundaries.
- AI Cosmos: Is Getting It Right with AI the Same as Learning It?
Use AI Cosmos's legal and programming studies to examine situated participation, executable workbenches, and the difference between earning points and mastering a target skill.
- Coding Agents: Why Does Permission to Fix Code Not Authorize Accounts and Network Access?
A fictional workflow separates task approval, tool execution, credentials, outbound data, and audit evidence to explain why coding agents need several enforceable boundaries.
- OpenGray: Why Decision Accuracy Above 90% Can Hide Failures to Stop
Use OpenGray's known-answer tasks and planning logs to distinguish aggregate decision scores, escalation, dose-goal attainment, and deliverability.
- Panoptik: When Does an AI Video Edit Actually Wait for Your Approval?
Panoptik shows why structured video edits are useful to inspect—and why a documented human-approval workflow must be checked against the paths that actually write to the timeline.
- GravityOCR: Why 3.94× Faster Decoding Becomes a 1.32× Page Speedup
Understand GravityOCR's shared diffusion drafter and AR verifier through three timing boundaries, output-equivalence conditions, and OCR pipeline bottlenecks.
- ABC-LLM: How Can AI Edit a Passage and Preserve the Rest of the Score?
How ABC states, operation scopes, and preservation conditions turn music generation into incremental editing—and what the evaluation actually establishes.
- Hy-Nomas: remembering scenes while letting actions end
How action timing, Prompt Enhancer and video history work together in Hy-Nomas, and what human comparisons, spatial memory and deployment conditions establish.
Huahua checked every note
Nothing matches this search yet.
Try a broader keyword, switch the content type, or continue from one of the curated paths below.