Project ideas from Hacker News discussions.

Humanising LLM Outputs Is Dumb

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

3 Dominant Themes in the Discussion

Theme	Supporting Quotations
1. Preference for terse, machine‑like output	“I don't get it. The skills and instruction try to make the answer more machine like on purpose.” – mikaeluman “Answer impersonally, objectively and analytically, without undue friendliness or enthusiasm. Use an engineering style response: concise, factual, and complete.” – prymitive
2. Critique of over‑personalisation and lock‑in tactics	“Ultimately a vendor lock‑in strategy, you'll get the best results with agents from the same tribe, others will trip over the mountain of idiosyncratic metaphors.” – whythismatters “People want it to be Data from Star Trek, when it really should be the ship's computer.” – MSFT_Edging
3. Need for simpler intermediate translation / two‑step summarisation	“Seems like something fixable with a simple two step process. Ask it the thing. Then ask it to summarise the answer in simpler terms.” – Havoc “pretty much what I do, better yet ask it to boil it down in visuals in a simple webpage if it's a very large project.” – kuberwastaken

🚀 Project Ideas

Generating project ideas…

Context Decompressor

Summary

Converts verbose LLM outputs into concise, structured digests (bullets, mindmaps, visual diagrams) to eliminate parsing fatigue.
Provides rapid understanding without re‑reading the original text.

Details

Key	Value
Target Audience	Developers, researchers, analysts who consume long AI‑generated text
Core Feature	Automatic summarization + visual mindmap generation
Tech Stack	Node.js backend, React front‑end, GPT‑4 API for parsing, Mermaid/Chart.js for diagrams
Difficulty	Medium
Monetization	Revenue-ready: subscription tier $9/mo or annual $90

Notes

HN users repeatedly cited “no clue what you actually read” and wanting “simple docs” (e.g., “boil it down in visuals in a simple webpage”).
Could integrate with IDEs and research note‑taking apps, creating a clear marketplace for sharing digests.

Prompt Translator

Summary

Transforms AI‑generated verbose language into terse, domain‑specific instructions while preserving the underlying reasoning steps.
Enables users to keep the “semantic work” unchanged but receive a clean output format.

Details

Key	Value
Target Audience	Engineers, product managers, power users of LLMs
Core Feature	Two‑step pipeline: generate raw answer → rewrite in user‑selected style (code‑like, bullet, LaTeX) without losing internal logic
Tech Stack	Python microservice, LangChain for internal chain, OpenAI GPT‑4 for rewriting, FastAPI wrapper
Difficulty	High
Monetization	Revenue-ready: pay‑as‑you‑go $0.01 per 1k tokens, tiered plans

Notes

Discussion points: “Ask it the thing. Then ask it to summarise the answer in simpler terms” and “keep the internal process intact up to the point of formulation.” Users want a tool that does exactly that.

Objective AI Associate

Summary

Strips away anthropomorphic language, personalization, and emotional signaling from LLM responses.
Provides a configurable “tool mode” that outputs only factual, engineering‑style text.

Details

Key	Value
Target Audience	Technical users, researchers, developers who treat LLMs as utilities
Core Feature	Toggle for “tool‑only” mode: no names, no hype, no emojis, pure objective output
Tech Stack	CLI wrapper around open‑source LLMs (e.g., Llama 3), custom prompt templates, Docker container for easy deployment
Difficulty	Low
Monetization	Hobby

Notes

HN comments lament “pretending it’s a conscious being” and desire “Data from Star Trek” vs “ship’s computer”. This product satisfies the need for a strictly functional assistant.