IdeaWell

Project ideas from Hacker News discussions.

Claude reveals ECDSA curve origins

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Four Prevalent Themes in the Discussion

1. Discovery of NIST Curve Seed Origins via Claude

The discussion centered on how Anthropic's Claude model revealed that NIST elliptic curve seeds (P-192, P-224, etc.) originated from lrand48() seeded with Unix timestamps—not from hashed phrases or intentional backdoors—putting long-standing conspiracy theories to rest.

JumpCrisscross: "In under half an hour, Claude discovered the origins of the NIST P-192, P-224, P-256, P-384 and P-521 seeds. None of them were the results of any hash algorithm at all, but rather the output of the C library’s lrand48 random-number generator, seeded with a Unix timestamp!" sweis: "I think they legitimately did not know. The person who worked on it directly, Jerry Solinas, did not remember and gave an explanation that turned out to be wrong." tptacek: "Nobody conversant with elliptic curve cryptography research believed there was a backdoor based on these seeds; for a number of reasons, best articulated by Koblitz and Menezes..."

2. Anthropic's Restrictive Access Policies (Cyber Verification Program)

Users frequently criticized Anthropic for blocking or downgrading legitimate cryptography, security, and biology research for ordinary users while allowing internal teams to conduct such work freely, creating frustration about unequal access.

slowin: "Anthropic (and whoever Anthropic wants) can do this kind of work with Claude, but if you or I try to do it, we will get blocked." petesergeant: "Fable 5.1 has downgraded me in the past week alone for the crimes of describing a coordinating server as 'the C&C server' ([cyber]), for doing some statistical analysis on some toy Vigenère-ciphered text ([cyber]), and for asking it to help me find patterns in some semi-random text ([bio])" AtNightWeCode: "For me it refused to move Claude transcripts from point A to point B on my local machine. It kept getting flagged as reasoning extraction or something."

3. Concerns About AI Concentration of Power and Regulatory Capture

Commenters warned that frontier AI labs are creating a dangerous "moat" by hoarding capabilities, influencing policy to restrict competition, and positioning themselves as gatekeepers of scientific progress—potentially stifling open innovation.

richwater: "It doesn't matter what is blocked today, it matters what is blocked tomorrow. If we allow these providers to promote regulatory capture they become the arbiter of what anyone is 'allowed' to do" Aachen: "Which brings us back to the war on general computation. So long as people can have computers that they control, at least in one of the countries or regions, there's no putting this cat back in the bag..." overfeed: "It's hilarious how a the trillions invested in AI companies, and their valuations are predicated on the AI labs Sherlocking entire industries, and yet CEOs, mathematicians, and everyone else are eagerly feeding the beast their most prized information."

4. Debate Over Justification for AI Capability Restrictions

The thread split on whether limiting AI access (e.g., for harmful information like weapon recipes) is necessary for safety or an overreach that hinders legitimate research, with arguments about real-world barriers to misuse versus philosophical principles of open knowledge.

solyin: "Because there are good and bad people in the world, and if I ask Claude 'how do I make cocaine' and it gives me a thorough recipe, it doesn't know if I'm being intellectually curious or trying to become the next Pablo Escobar." hgoel: "Biologists already addressed this idiocy a few weeks ago [1]. Making a biological weapon requires facilities that only some nation states and comparable companies, have access to. AI isn't going to magically assemble a DNA sequence for you..." jiggawatts: "Not to mention that making an actual novel virus requires thousands of dollars of equipment, serious lab skills, and even then is so dangerous that the wannabe terrorist is more likely to infect and kill themselves than cause mass causalities."

🚀 Project Ideas

Open CryptoReasoner

Summary

  • A fine‑tuned, openly licensed LLM optimized for cryptographic and historical research tasks (e.g., tracing NIST curve seeds, analyzing PRNG outputs).
  • Provides researchers with a self‑hostable model that can be prompted without the restrictive “Cyber Verification” gatekeeping, enabling reproducible investigations like the Claude‑solved seed origin bounty.

Details

Key Value
Target Audience Cryptographers, security researchers, academia, independent investigators
Core Feature Open‑weight model (e.g., Llama‑3‑70B) fine‑tuned on crypto‑literature, PRNG analysis, and prompt‑guided reasoning chains; includes a curated prompt library for common investigative tasks
Tech Stack Python, HuggingFace Transformers, bitsandbytes for quantization, Docker, optional GPU‑accelerated inference via vLLM
Difficulty Medium
Monetization Hobby

Notes

  • HN commenters expressed frustration that “Anthropic’s cyber program blocks normal people from doing crypto work” (novia) and desire for “open weights … no elite gatekeepers” (verdverm).
  • By offering a fully open model with transparent licensing, researchers can run the same kind of reasoning locally, verify claims, and share prompts without needing corporate approval.

ReasonTrace Verifier

Summary

  • A web platform where users can upload LLM internal reasoning traces (chain‑of‑thought logs) and receive community‑driven validation, cryptographic timestamps, and annotation tools.
  • Addresses the need for trustworthy AI reasoning, letting skeptics inspect whether a model truly arrived at a conclusion (e.g., the seed origin) or merely hallucinated.

Details

Key Value
Target Audience AI safety researchers, developers, journalists, anyone scrutinizing LLM outputs
Core Feature Trace upload → automatic parsing, diff‑against‑prompt, community voting, optional zero‑knowledge proof of trace integrity; integrates with model APIs to fetch traces when available
Tech Stack React/Vue frontend, Node.js backend, PostgreSQL, IPFS for storage, optionally Circom/ZK‑SNARKs for proof generation
Difficulty Medium
Monetization Hobby

Notes

  • Users asked for visibility into Claude’s internal reasoning (“the hmm… in the internal reasoning process log”) and wanted to know if the model truly performed the work (nullc, tptacek).
  • A verifiable trace repository would let the HN community independently assess claims, echoing the call for “open weights” and transparency.

FOIAraph

Summary

  • An automated FOIA request generator and tracker tailored to U.S. government agencies (NIST, NSA, etc.), offering pre‑built templates for common research topics and a dashboard to monitor status, deadlines, and released documents.
  • Solves the pain point of manually crafting and following up on FOIA requests, which several commenters found cumbersome (sweis, hoten).

Details

Key Value
Target Audience Researchers, journalists, hobby investigators needing government records
Core Feature Template library (e.g., “origin of NIST curve seeds”, “Dual_EC_DRBG documentation”), AI‑assisted request drafting, automatic submission via agency portals or email, status notifications, document OCR/search
Tech Stack Next.js, TailwindCSS, Python FastAPI, AWS SES/S3 for attachments, OAuth for agency portals, optional CAPTCHA‑solving microservice
Difficulty Low
Monetization Hobby

Notes

  • Commenters noted difficulties filing FOIA requests (“I don’t know how to write these FOIA requests in the right way”) and expressed interest in obtaining primary source proof (sweis, hoten).
  • FOIAraph lowers the barrier, enabling more public scrutiny of agency actions—a topic that resonated with HN’s focus on accountability.

ResearchNotebook AI

Summary

  • An open‑source, Jupyter‑like notebook that tightly integrates a locally run LLM, web search, code execution, and version‑controlled prompts, allowing users to reproduce end‑to‑end investigations (like the NIST seed bounty) with full provenance.
  • Provides a reproducible research environment that mitigates reliance on external, restricted model APIs.

Details

Key Value
Target Audience Researchers, students, independent investigators conducting technical or historical inquiries
Core Feature Notebook cells that can call a local LLM (via Ollama/LlamaCpp), fetch URLs, run Python/R/Julia code, and store prompt‑output pairs in a Git‑backed notebook; includes replay and diff tools
Tech Stack Electron or Tauri desktop app, Monaco Editor, Ollama/LlamaCpp backend, SQLite for metadata, Git integration
Difficulty Medium
Monetization Hobby

Notes

  • Several HN users lamented that “we used a machine to solve it at all” and wished humans could have done the work themselves (Minraws).
  • ResearchNotebook AI lets users run the same LLM locally, inspect every step, and share the exact notebook for others to verify—directly addressing the desire for reproducible, transparent AI‑assisted research.

GPUShare

Summary

  • A peer‑to‑peer compute marketplace where individuals can rent GPU time to run unrestricted, large‑scale LLMs for research, bypassing provider‑imposed usage limits.
  • Includes reputation scoring, escrow payments, and optional pre‑loaded crypto‑research model images to get started quickly.

Details

Key Value
Target Audience Developers, researchers, small teams needing powerful AI compute without corporate gatekeeping
Core Feature Marketplace matching GPU providers with consumers; Docker images pre‑loaded with open LLMs (e.g., Mixtral, Llama‑3) and research tools; smart‑contract‑based escrow and reputation system
Tech Stack Web app (Svelte), Node.js/microservices, WebRTC for direct streaming (optional), Ethereum L2 or Solana for payments, IPFS for image storage
Difficulty High
Monetization Revenue-ready: {transaction fee ~5% per rental}

Notes

  • Commenters warned of a “moat widening between model providers, their government partners, and the public” (slowin) and wanted “open models” to avoid restrictions (A_D_E_P_T).
  • GPUShare gives the public direct access to compute, letting them run the same class of models that labs use internally, thus reducing the perceived power imbalance.

Read Later