> Software engineers who use LLMs should save a curated prose document describing the software, alongside the code that was generated from it.
That sounds great! Though software engineers are all about being lazy - typically in a good way. So I expect in practice they'll use LLMs to generate some code, iterate a few times until it works well enough for them, and then have the LLM generate "a curated prose document describing the software..."
It doesn't sound great to me. That would be a bit like if I asked my contractor to read a 500 page tome instead when he wanted to come look at the studs in the walls, because the original architect made an incomprehensible mess of it.
Thanks! I've heard numerous examples of this. E.g. paint rollers, and later paint sprayers, made painting so much more efficient and cheaper that people wanted their houses and apartments painted more often, ultimately increasing house painter employment. But I didn't realize the phenomenon had a name.
The broader topic of marginal utility is worth looking into if this is interesting to you.
The reason why AI induces more demand is more interesting than simply observing a correlation. At the root of the paradox is the dissatisfaction with AI as it raises the bar for human work. In that sense, AI functions like training wheels more than actually doing anything useful on its own. More people can ride the bike well as a result.
This is the "law of diminishing marginal utility". In other words, LLMs are an NLP-powered google search 2.0. That sums up pretty well why AI hit a plateau a couple of years ago. There's only so much training data and humans become more efficient and creative once exposed to the exact information they were always looking for, but could never find in the past.
Downloading and redistributing companies' exact AI models would violate copyright. Querying multiple AI models to form your own model with a unique set of weights does not.
Exactly. AI companies distilled the whole internet into AI weights, now other AI companies are distilling their work to generate new AI weights. Nothing wrong with that, and the world gets open weights AI models.
That sounds great! Though software engineers are all about being lazy - typically in a good way. So I expect in practice they'll use LLMs to generate some code, iterate a few times until it works well enough for them, and then have the LLM generate "a curated prose document describing the software..."
reply