Project ideas from Hacker News discussions.

Schema Harness Achieves ~99% on Arc‑AGI‑3 Public

📝 Discussion Summary (Click to expand)

Three dominant themes

Theme Summary Supporting quotation
1. Public‑set scores are misleading Many commenters stress that a 99 % result on the public ARC‑AGI‑3 set does not imply real generalisation. “It doesn't necessarily mean anything to reach 99% score on Arc‑AGI‑3. All of the public set is known in advance, so it's possible to hardcode rules that make this easy for the models.” – modeless
2. The harness sidesteps the test’s intent Several users argue that the presented method essentially “cheats” by building a game simulator and planning instead of letting the model infer rules organically. “You go to get your chess ELO. You don't know chess at all… you pull out your laptop and write a chess engine. Then when you go to get ranked, you just copy the moves from the software. Now you're a grand master.” – Tadpole9181
3. Mechanistic discovery as a genuine breakthrough Despite the criticism, a number of participants see value in the approach of converting game observations into an executable world model, labeling it a meaningful step toward agency‑level AI. “State grounding turns raw observations into objects, variables, and relations that can be tracked. Mechanism discovery finds how that state changes under an action and writes the rule as an executable program.” – ClassAndBurn

These three threads dominate the discussion: skepticism about benchmark results, concerns that the evaluation is being gamed, and tentative praise for the underlying mechanistic‑discovery technique.


🚀 Project Ideas

[ARC Benchmark Transparency Hub]

Summary

  • A public, auditable platform that enforces open‑source harnesses and validates private‑set scores for ARC‑style reasoning benchmarks.
  • Eliminates “harness cheating” by requiring transparent submission and independent verification of results.

Details

Key Value
Target Audience AI researchers, model developers, ARC‑AGI participants
Core Feature Mandatory open‑source harnesses + verified private‑set scoring
Tech Stack React front‑end, FastAPI backend, Docker, CI/CD pipelines
Difficulty Medium
Monetization Revenue-ready: subscription $29/mo per team

Notes

  • HN commenters repeatedly asked for proof of private‑set performance; this directly answers that demand.
  • Provides a community‑driven marketplace for vetted harnesses, increasing trust and reducing debate.

[Simulator‑Guard Framework]

Summary

  • A toolkit that guides LLM agents to emit interpretable rule‑programs rather than full game copies, with built‑in verification of state transitions.
  • Prevents superficial “simulator” hacks while preserving the benefits of mechanism discovery.

Details

Key Value
Target Audience AI product teams building rule‑discovery agents for puzzles or games
Core Feature Generation of constrained executable rule modules + automated correctness checks
Tech Stack Python, LangChain, PyTorch, FastAPI, Docker
Difficulty High
Monetization Revenue-ready: usage‑based $0.01 per inference

Notes

  • Addresses concerns that current harnesses merely let models “cheat” by building simulators.
  • HN users emphasized the need for genuine rule inference; this framework delivers it.

[Mechanism Discovery Studio]

Summary

  • SaaS platform offering template pipelines that transform raw observations into causal models and executable rule programs for complex environments.
  • Lowers the barrier for teams to adopt mechanism discovery without deep expertise.

Details

Key Value
Target Audience Startups and research labs focused on generalist AI and robust rule learning
Core Feature End‑to‑end pipeline (perception → state → dynamics → executable rule) with fine‑tuning UI
Tech Stack Rust backend, ONNX models, Streamlit UI, Kubernetes
Difficulty High
Monetization Revenue-ready: tiered SaaS $199–$799 per month

Notes

  • HN discussants highlighted the potential of “mechanism discovery” for real‑world messy environments.
  • Provides a concrete service that turns that potential into an actionable workflow.

Read Later