{
  "id": 16395170,
  "title": "My AI Lost to Simple Rules. So I Built BenchSloth Around Evidence.",
  "url": "https://urgent.news/2026/10/25/my-ai-lost-to-simple-rules-so-i-built-benchsloth-around-evidence",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-25T00:10:11.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/siamakenna/my-ai-lost-to-simple-rules-so-i-built-benchsloth-around-evidence-5966"
  },
  "original_language": "en",
  "account": "This submission for the Hacktoberfest Open-Source AI Challenge focuses on the concept of \"Slow and Steady.\" The author discovered that limiting the model's authority led to better results. The AI policy provided only three eligible waits out of 13 options, while guarded rules offered ten. The author realized that giving the AI too much control was unnecessary and instead built an agent that managed attention around evidence.\n\nBenchSloth is the result of this idea. It provides a plan for attention rather than just an experiment. The tool includes three main parts: Source, Attention, and Intelligence. The Source keeps the original words and occurrence IDs intact, while the Attention shows fixed intervals and review needs in source order. The Intelligence section allows for inspecting evidence selected for a question.\n\nWhen using BenchSloth, users can ask which earlier wait a note describes, open quoted passages, inspect the actual model request, record corrections, confirm intervals, and return to the next instruction. The agent records the path between these actions. However, it does not claim to have watched or completed the experiment.\n\nThe demo starts with the clear wait (A), which includes a Gemma response, measured duration, and verification trace. The uncertain note (B) maintains the distinction between two identical waits, and the longer workflow (C) consists of thirteen lines across two sections. By returning to the initial wait (A) and confirming its ten-second demonstration, users can breathe easier knowing that the clock and source evidence remain visible.\n\nThe source material is preserved in the three parts of BenchSloth, and the agent's path is deterministic. If the source lacks an exact duration or relevant target, the tool cannot grant a timer. BenchSloth successfully recovered 22 out of 22 reference links and 16 out of 18 eligible waits in synthetic cases.",
  "summary": "This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 3 : Slow and Steady. The most useful result in my AI benchmark was the one telling me to give the model less authority. My first lab-note assistant recognized a wait—and then made it harder to use. Its model-driven policy offered just 3 of 13 eligible waits . Guarded rules offered 10 . The clever part was getting in the way…",
  "key_points": [
    "BenchSloth simplifies AI decision-making by focusing on evidence-based attention management.",
    "The tool offers three components: Source, Attention, and Intelligence for structured AI interaction."
  ],
  "editors_take": "Prioritizing evidence-based attention over model autonomy shifts the focus from raw intelligence to verifiable accuracy, providing a structured framework for users to audit and validate AI-generated reasoning.",
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}