Drawing a coffee plot, with LLMs ep0

LLM-assisted work?

Will the LLM take over us? Yes, no. Maybe.

The only thing I’m sure about is that those cool stuff are not able to assess the certainty of text (“opinions?”) they generate.

Moreover, In my humanly-limited experience, I have yet to see LLMs to meaningful contribute to any asset, i.e. to add valuable and extensive contribution to non-throaway code. All of this may seem strong, but hey, this does not mean productivity cannot be positively impacted.

Today, I’ll be impersonating a journalist or blogger, dealing with a niche site generator as a contraint [^1]. I need to re-elaborate some data from EFSA about coffee in 5 minutes.

Rules of the game, for today

Here are the rules of the game I’ll start with:

  • On the D3.js part, I’ll be as “clean hands” as possible, avoiding any major code contribution
  • As anticipated, the site generator works as a “constraint”, i.e. the plot has to be integrated with an existing system (this blog)
  • The initial idea of the design is provided very clearly (a pre-made plot, from EFSA),
  • I am evaluating only the “graphic” capabilities, for the time being.

You can see below how it went, with actual reference, actually it took more than 5 minutes (about 15).

Day 0: result

Done with

Honestly, it is good, but it took about 3x what I wanted to. Let’s see what can be achieved with anoter round. After all, estimation is a long-standing issue in software development.

[^1] it’s called Hugo. I like it actually, but for the sake of the exercise it a “contraint” applied on top of the LLM

comments powered by Disqus