IdeaWell

Project ideas from Hacker News discussions.

Show HN: How Random Are You?

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Theme 1 – Privacy/Paranoia about data collection
- pavel_lishin: “Is anyone sufficiently paranoid that their first thought was, ‘Are you going to train a model on my voice?’”
- progforlyfe: “Yes, my mind went directly here.”

Theme 2 – Human number generation is predictable (Aaronson‑oracle effect)
- costco: “I wrote a simple little program that would let people type either “f” or “d” and would predict which key they were going to push next… Even a very crude one will do well… I couldn’t even beat my own program… Then, we found one student that the program predicted exactly 50% of the time.”
- vorticalbox: “I got to 2/30 then it just guessed 7 from 30‑50. Did I break it?”

Theme 3 – Users notice the system’s guesses and play adversarially, finding it fun
- alwa: “I found it interesting when I felt my focus shift from the number‑generating task to noticing (and responding to) the house’s guesses. As soon as I noticed it repeating its guess, I had a good long run of adversarially avoiding that number… it’s kind of fun being on the machine side of the slot‑machine‑type racket!”
- vorticalbox (implicitly): his attempt to “break” the predictor shows the same cat‑and‑mouse dynamic.

🚀 Project Ideas

Generating project ideas…

Keystroke Predictability Lab – Local‑First

Summary

  • A client‑side web app that lets users type “f” or “d” (or speak) and see a live prediction model run entirely in the browser, guaranteeing no data leaves the device.
  • Core value proposition: gives users transparency and control over their data while letting them explore predictability and privacy concerns.

Details

Key Value
Target Audience Privacy‑conscious developers, hobbyists, students interested in ML and human behavior
Core Feature Real‑time, in‑browser prediction model (e.g., TensorFlow.js) that updates as the user types, with stats display and reset option
Tech Stack HTML/CSS, JavaScript, TensorFlow.js (or ONNX Runtime Web), optional Web Speech API for voice
Difficulty Medium
Monetization Hobby

Notes

  • HN commenters voiced paranoia about voice training (“Are you going to train a model on my voice?”) – this tool runs locally, directly addressing that fear.
  • Can spark discussion about predictability, free will, and serve as an educational demo for ML courses.

Predictor Challenger – Beat the AI

Summary

  • An interactive game where users try to avoid being predicted by an adaptive model that learns from their past inputs (f/d, numbers, or gestures).
  • Core value proposition: turns the Aaronson oracle concept into a fun, competitive challenge with scoring and leaderboards.

Details

Key Value
Target Audience Gamers, puzzle enthusiasts, researchers studying human randomness
Core Feature Adaptive prediction model (Markov chain or small neural net) that updates after each move, shows prediction confidence, and awards points for evading prediction
Tech Stack React frontend, Node.js/Express backend (optional for leaderboard), Socket.IO for real‑time, Python/TensorFlow.js for model
Difficulty Medium
Monetization Revenue-ready: Freemium with optional premium skins or ad‑free experience

Notes

  • Users described fun adversarial avoidance (“As soon as I noticed it repeating its guess, I had a good long run of adversarially avoiding that number…”) – this builds on that enjoyment.
  • Could lead to discussions about randomness, strategy, and ML; useful for behavioral experiments.

HumanLikeSeq – Synthetic Human Input Generator

Summary

  • A library/service that generates realistic human‑like sequences (keystrokes, voice transcripts, mouse movements) using models trained on public audio/podcast corpora, for testing anti‑fraud, CAPTCHA, and usability systems.
  • Core value proposition: provides developers with reproducible, controllable synthetic data that mimics human predictability without privacy concerns.

Details

Key Value
Target Audience QA engineers, security researchers, product teams building bot detection
Core Feature Configurable generator outputting sequences with adjustable entropy, pattern strength, and noise; includes export to CSV/JSON
Tech Stack Python (PyTorch/TensorFlow), optional FastAPI for REST API, Docker for deployment
Difficulty High
Monetization Revenue-ready: Subscription: $0.01 per 1k sequences

Notes

  • Commenters noted the abundance of radio/podcast data (“Why would they when there are thousands of hours of radio and podcasts do train on?”) – this leverages that resource ethically.
  • Enables discussion about data sourcing, model bias, and practical utility for testing anti‑abuse systems.

Read Later