Post #1163429

2026-04-14 03:15 UTC

One thing I haven&#39;t heard discussed much about LLM-based programming: Are we at the end of any serious programming language evolution? It seems to me that if you believe (a) coding assistants are going to be the dominant way software is written, and (b) the only way to get reasonably good coding assistants is to train on tons of existing code, then it follows that (c) new programming languages will be at a significant disadvantage because there&#39;s just no corpus to train on. The strong version of this hypothesis is that even new features of existing programming languages will be hard to roll out, because the bots won&#39;t use them! I&#39;m sure there are researchers who are looking at &quot;programming languages that are easy for coding assistants to write&quot; (Darklang went this route) but I think this is futile as long as (b) is our best bet. Maybe &quot;AI&quot; will deliver general reasoning capabilities good enough to transfer to any language, but that&#39;s not what is shaking out right now. Any studies on coding assistant success across existing languages? That might be interesting to look at. #CodingAssistants #ProgrammingLanguages

Replies (5)

@markgritter@mathstodon.xyz 2026-04-14 04:36

This seems vaguely familiar, but I don't think I've read it yet: https://arxiv.org/abs/2603.09678v1 "We evaluate five frontier models across five prompting strategies and find a dramatic capability gap: models achieving 85-95% on standard benchmarks score only 0-11% on equivalent esoteric tasks, with 0% accuracy beyond the Easy tier"

Open ##1472677
@skewray@mathstodon.xyz 2026-04-14 04:18

@markgritter You make a very strong argument that what we need is a new programming language that doesn't allow posting code where it can be scanned by demons.

Open ##1472678
@gray17@mastodon.social 2026-04-14 06:16

@markgritter I've been idly thinking about LLM-hostile programming languages. the tricky part is making it ergonomic. the ideas I've had so far are meh

Open ##1472686
@jaseg@chaos.social 2026-04-14 06:38

@markgritter I think this stagnation effect is similar in nature, but stronger compared to the split between “hip” languages like js and rust, and languages extensively used in large-scale business use cases and in introductional CS teaching at unis like Java, C/C++ and increasingly python.

Open ##1472687
@poleguy@mastodon.social 2026-04-14 07:21

@markgritter Is (b) true? I'd think it would do fine with the grammar of the new language and a good codebase in any other language. I'd imagine this could even be done just within the context window without even re-training the LLM. But I don't actually know.

Open ##1472689