FRESH

Hacker News

Home

This is the future: A Software Library with No Code

30 points by ulrischa

by roxolotl

1 subcomments

Why wouldn't we just specify these things with programming languages though? The purpose of a programming language is to enable precise specification of computation. SPEC.md is 497 lines, the tests another 500. I don't understand what we'd gain from taking a spec which is imprecise and tests, which seemingly are arbitrary and cover the same cases multiple times, putting them into an LLM and recreating a version of the computation you desire.
If an LLM can do this with English and arbitrary test cases why wouldn't you pick a programming language and specific test cases? This would give you significantly more repeatability and consistency all while probably having less overall lines of code.

by pseudohadamard

0 subcomment

>whenwords provides five functions that convert between timestamps and human-readable strings, like turning a UNIX timestamp into “3 hours ago”.
Ah, so you're the person responsible for this brain rot. Could you please DM me your name and address? I think the 20-to-life I'll get will be well worth it.

by simonw

0 subcomment

I've been exploring this pattern recently too. Giving current coding agents an existing conformance or test suite and telling them to keep writing code unto the tests pass is astonishingly effective.
I've now got a JavaScript interpreter and a WebAssembly runtime written in Python, built by Claude Code for web run from my phone.

by NitpickLawyer

0 subcomment

Ok, I really liked this! A cool concept that would've been sci-fi ~5 years ago (not this specific library, but the idea of spec->code in any language, 0 human intervention). I can see a collection of utilities "specced out" and then shared as specs + tests. Perhaps for harder stuff even some spec formal verification + tests? (no idea how feasible this is, but it's something to look into)
I had no doubts claude/codex/gemini could handle this, but I was surprised that even an open-source model (glm4.7) did it 1 shot (just c/p the prompt from Install.md and came back 10 min later):
============================= 123 passed in 0.18s ==============================
Test Results:
```
  All 123 tests from tests.yaml pass successfully:
  - 35 timeago tests
  - 26 duration tests
  - 28 parse_duration tests
  - 19 human_date tests
  - 9 date_range tests

  Files Created:

  1. /workspaces/glmcode/whenwords/src/__init__.py - Library implementation
  2. /workspaces/glmcode/whenwords/test_whenwords.py - Test file generated from tests.yaml
  3. /workspaces/glmcode/whenwords/usage.md - Usage documentation

 Crunched for 11m 38s
```
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── Showing detailed transcript · ctrl+o to toggle
I mean, it's a "toy" library, but the concept is so cool! And the fact that an open, locally hostable model can do it 1shot is insane.