IdeaWell

Project ideas from Hacker News discussions.

How do we prevent mathemathics from devolving into the Medieval Era of secrecy?

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Three prevalent themes in the discussion

  1. Need for legal/explicit protections against AI training on mathematicians’ work
    Many commenters argue that researchers must prevent AI companies from freely using their publications for model training.

    “Mathematicians/Scientists/Researchers need to stop sharing freely with ‘AI Companies’ and have explicit clauses in place in their publications about not using their research without their explicit consent.” – rramadass
    “Just putting a clause in a publication won't prevent it from being used as training data.” – nradov
    “Information wants to be free” never meant that people want to release their information; it meant that information is very hard to keep secret… – pavel_lishin

  2. Tension between openness (“information wants to be free”) and the growing pressure toward secrecy
    The debate centers on whether the traditional ethos of open sharing is eroding because AI can instantly appropriate ideas, prompting calls for either earlier publishing to stake claims or increased secrecy.

    “I don't like this and I wish it weren't true, but I think the period of 'information wants to be free' is coming to an end…” – Analemma_
    “This is about cooperation before publishing results. And they will keep everything medieval secret, else some big company steals it and claims it their own.” – watwut
    “Two people can keep a secret if one of them is dead.” – nradov

  3. Institutional and technical responses: private LLMs, specialized tools, and policy changes
    Several participants propose that universities or labs host their own models, restrict use of public AI services, and adopt purpose‑built tools to retain control over their work.

    “I'd guess universities might starting hosting open source models. They can probably actually afford to, unlike individual mathematicians.” – perching_aix
    “Universities should be providing university hosted llms to their faculty and students…. It is going to have to be against the rules to submit anyone else's work to a commercial llm.” – roosterIllusi0n
    “Mathematicians should be using sophisticated, specialized LLM tools in much more sophisticated ways than lay people.” – mmooss

🚀 Project Ideas

AI-Train Consent Ledger

Summary

  • A decentralized registry where researchers can attach machine‑readable AI usage permissions (opt‑in, opt‑out, or specific conditions) to each publication or dataset.
  • Provides publishers and LLM crawlers with a standardized way to respect authors’ wishes, reducing unauthorized training on scholarly work.

Details

Key Value
Target Audience Mathematicians, scientists, academic publishers, AI labs
Core Feature Blockchain‑based ledger storing hashed metadata + usage license; API for validators to check permissions before crawling
Tech Stack Ethereum/IPFS (or Polygon for low cost), Solidity smart contracts, Python/Flask backend, React frontend, OAuth2 for researcher identity
Difficulty Medium
Monetization Revenue-ready: Subscription tier for institutions + per‑record gas fee rebate

Notes

  • Addresses rramadass’s call for “a clear legal distinction between using research data for AI model‑training vs another researcher using it” by giving authors a machine‑readable way to state that distinction.
  • Enables roosterIllusi0n’s desire for universities to control model access: registries can be queried by campus LLMs to block training on opted‑out works, giving practical utility to the discussion.

CampusLLM Hub

Summary

  • A turnkey platform that lets universities deploy and manage private, open‑weight LLMs (e.g., Llama, Mistral) with role‑based access, audit logging, and usage policies.
  • Ensures faculty and students can experiment with LLMs without exposing their data to commercial services.

Details

Key Value
Target Audience University IT departments, research labs, graduate students
Core Feature One‑click Kubernetes deployment of LLMs with SSO/LDAP integration, usage quotas, and opt‑out/training‑data isolation
Tech Stack Kubernetes, Helm charts, HuggingFace Transformers, Triton Inference Server, Keycloak for auth, Grafana/Prometheus for monitoring
Difficulty High
Monetization Revenue-ready: Tiered pricing based on GPU‑hours consumed + optional support contracts

Notes

  • Directly satisfies roosterIllusi0n’s comment: “Universities should be providing university hosted llms to their faculty and students.”
  • Responds to nradov’s suggestion that scholars either purchase enterprise licenses or run private instances, offering a cheaper, centrally managed alternative.

MathProof Guard

Summary

  • A service that embeds cryptographically‑secure, invisible watermarks into LaTeX/PDF manuscripts and monitors public LLM outputs for matches, alerting authors when their work appears in model‑generated text.
  • Gives researchers concrete evidence of unauthorized AI training use and supports enforcement of usage policies.

Details

Key Value
Target Audience Individual researchers, research groups, academic publishers
Core Feature Watermark embedding tool (CLI/Zotero plugin) + similarity‑search engine scanning LLM APIs or public model checkpoints for leaked content
Tech Stack Python (pylatexenc, hashlib), FAISS for vector similarity, Docker, optional GPU for fast inference, web dashboard (React)
Difficulty Medium
Monetization Revenue-ready: Pay‑per‑document watermark + annual monitoring subscription

Notes

  • Echoes antonvs’s worry about “the existence of the triplets NSA/CIA/GRU” and the need to protect intellectual property, offering a technical means to detect misuse.
  • Provides a practical tool for the “legal framework” discussion, letting mathematicians prove when their work is used without consent, which could motivate the enforcement mechanisms users called for.

Read Later