IdeaWell

Project ideas from Hacker News discussions.

Bend 2 and the Vibe-Coding Trap

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

1. Vibe‑coding / AI‑generated code concerns
Many commenters argued that Bend’s compiler being “99% AI‑written” (as stated in its README) exemplifies unvetted, “vibe‑coded” slop that lacks proper auditing or understanding of the field.

“The compiler (not kernel) is 99% AI-written and has not been fully audited yet.” – wg0 (quoting the README)
“so vibecoded” – jdiaz97

2. Defense of the author’s expertise and intent
Opposing voices pointed out that the language’s creator, Victor Taelin, has a long‑standing background in formal verification (e.g., Formality, Cedille‑Core) and that the design choices are deliberate, not the result of ignorance.

“I dropped out of Federal University of Rio de Janeiro to study this subject independently… I’ve also been researching the later, and I built SupGen…”. – LightMachine (author)
“the author is clearly aware of formal verification, they've written several implementations of dependently typed languages”. – mccoyb

3. Technical debate over string representation and performance
Discussion centered on whether storing strings as linked lists (as Bend does) is a sensible trade‑off, especially for GPU‑parallel workloads versus cache‑friendly contiguous buffers.

“Note that using linked lists for strings is actually more 'parallel friendly' because you can take the head/tail and spread it around 16k GPU cores in O(1), unlike in Haskell, unlike arrays, which require a linear copy, becoming quadratic.” – LightMachine
“Strings are linked lists of characters, so text processing is slow.” – README.md (cited by wg0)

4. Skepticism about project popularity, star inflation, and repo history
Several users questioned the legitimacy of Bend’s rapid GitHub star growth, suggested possible botting, and criticized the decision to rewrite history when reusing the Bend1 repo for Bend2.

“The obvious botted stars on GitHub in 2 hours was an immediate red flag.” – rvz
“Does anyone know where they bought the popularity and contributors from?” – monster_truck
“Why would you kill the source history of Bend1 completely if it has 20K stars?” – bhouston

🚀 Project Ideas

LLM Prior Art Assistant

Summary

  • A retrieval‑augmented prompting tool that, before code generation, searches public codebases, academic papers, and language specs for existing solutions matching the user’s description.
  • Reduces vibe‑coding waste by surfacing prior art so LLMs can build on known work instead of reinventing.

Details

Key Value
Target Audience Developers using LLMs for code or language design, PL researchers, hobbyists experimenting with new features
Core Feature Given a natural‑language description, returns ranked prior art (libraries, algorithms, formal specs) and injects them into the LLM prompt as context
Tech Stack Sentence‑Transformers embeddings, FAISS vector store, periodic crawler of GitHub, GitLab, arXiv, PLDB; API wrapper (FastAPI) that can be called by Cursor, Copilot, or custom LLM scripts
Difficulty Medium
Monetization Revenue-ready: tiered SaaS subscription (free limited queries, paid for higher rate limits and private repo indexing)

Notes

  • HN commenters repeatedly asked for “run prior work research first” to avoid reinventing formal verification wheels (e.g., LiamPowell, SimonW). This tool directly satisfies that request.
  • Could spark discussion on how LLMs should be prompted to check existing literature, echoing the debate about AI‑slop and the need for better research‑first workflows.

Formal Verification Proof Optimizer

Summary

  • Takes verbose, LLM‑generated proofs (like those in Bend) and applies automated lemma extraction, rewriting, and calls to ATP/SMT solvers to produce compact, check‑friendly proofs.
  • Keeps the explicit‑proof design goal while dramatically reducing proof size and checking time.

Details

Key Value
Target Audience Language designers, proof engineers, teams using LLMs to generate correctness proofs (e.g., Bend, Lean, Coq users)
Core Feature Proof compression pipeline: parse proof → identify reusable sub‑proofs → replace with lemmas → optionally discharge with Z3/Vampire → output minimized proof
Tech Stack Rust for parsing Lean/Agda/Coq proof formats, rewriting library (egglog/equational‑reasoning), Z3/Vampire bindings, CLI and WASM wrapper for editor integration
Difficulty High
Monetization Hobby (open‑source) – could later evolve into a licensed plugin for CI verification pipelines

Notes

  • LightMachine argued that verbosity speeds up proof checking; this tool offers a best‑of‑both‑worlds approach: keep verbosity for the checker but hide it from humans via automated lemma extraction.
  • Addresses the critique that LLMs waste tokens on trivial proof steps (LiamPowell) while preserving the author’s intent of fast checking.

AI Slop Detector for Open Source Projects

Summary

  • A lightweight scoring service that analyzes a GitHub repo’s metadata (commit patterns, issue activity, star growth, README similarity) to estimate the likelihood of AI‑generated slop or low‑effort vibe‑coding.
  • Helps maintainers, investors, and HN readers gauge project authenticity quickly.

Details

Key Value
Target Audience Open‑source maintainers, VC scouts, HN readers, language‑project evaluators
Core Feature ML classifier outputting a “slop score” (0‑100) based on features: commit frequency vs. star spikes, issue/PR comment patterns, generated‑text detectors in README/CI configs
Tech Stack Python, GitHub GraphQL API, feature extraction (PyGithub), scikit‑learn model (GradientBoosting) or lightweight LLM‑based anomaly detector, simple Flask/Django API + React dashboard
Difficulty Medium
Monetization Hobby (free public API); optional paid tier for private repo scanning and higher‑resolution reports

Notes

  • Many commenters accused Bend of having purchased stars and being AI‑generated vibe‑code (monster_truck, nullbio). This tool would give an objective signal to calm or confirm such suspicions.
  • Could fuel discussion on the ethics of star‑farming and what signals truly indicate community interest vs. artificial hype.

Language Design Prior Art Checker

Summary

  • Web app where users draft a language feature (syntax, type rule, construct) and receive instant matches from existing languages, papers, and libraries with links to docs and implementations.
  • Prevents accidental reinvention by making prior art discoverable at the point of ideation.

Details

Key Value
Target Audience Language designers, PL researchers, educators, curious programmers experimenting with new language ideas
Core Feature Semantic search over a curated corpus of language specs (e.g., PLDB, Rosetta Code, academic PL papers) using embeddings; returns ranked matches with highlighted similarities
Tech Stack Embedding model (SBERT) over normalized spec texts, Vector DB (Pinecone/Weaviate), Next.js frontend, backend in Node/Express; periodic upload of new specs from arXiv and language repos
Difficulty Medium
Monetization Hobby – ad‑free community tool; optional donation or sponsorship model for hosting costs

Notes

  • The discussion highlighted that many features in Bend (e.g., explicit proofs, dependent types) already exist in Lean, Ada/SPARK, Idris; a quick lookup could have saved debate (SimonW, LiamPowell).
  • Encourages better-informed language design conversations on HN, reducing the “vibe‑coding a language without knowing the field” critique.

Read Later