The open source ecosystem is changing

[Jeff Squyres]

This is not a good portent for open source: https://dpc.pw/posts/i-dont-want-your-prs-anymore/

This is also not promising: https://malus.sh/

Example: see some open source software that you want to use, but don’t want to have to worry about that pesky GPL license that it has?  Use the clean room method with one LLM to document its behavior and then a 2nd LLM to implement from those specifications.  And possibly implement it in a wholly different language than the original.

Poof — original license is no longer an issue.

If you divide open source into a few different categories:

1. Corporate-maintained software
2. Community (not corporate) maintained software
3. Hobby software

I’d say that #1 may be affected the least from AI/LLMs.  Corporate-paid developers will still continue to develop (using AI/LLMs or not) and that software will continue to be developed and maintained.

#2 will likely (is already) being affected by AI/LLMs — the above two are good examples of the fallout.  There’s bunches of news stories about open source maintainers who were first being drowned by AI slop PRs, and are now being drowned in decent-quality AI-generated PRs.  Both result in the same thing: oodles of PRs that need to be reviewed (by unpaid volunteers).

#3 may not be affected much — but may also end up with oodles more AI slop than actual crafted software.

Just my $0.02.  It may be too simplistic for the complicated, nuanced overall “open source” ecosystem.  Any one else have opinions here?

-- 

{+} Jeff Squyres

[Deven Phillips]

Just my 2¢.

    When I run across something like this, I fully expect the "vibe coded" rewrite to be substantially inferior and would not use it. Additionally, if they change the license, it would no longer be the open source implementation, so another reason I would avoid it. My final thought is that I have seen extremely capable software engineers struggle to "rewrite" legacy software while not breaking functionality and it takes years of effort for any kind of complex system. The likelihood of this coming up with a valid, secure, reliable, and performant result is so low IMHO as to be laughable.

Deven

To unsubscribe from this group and stop receiving emails from it, send an email to kyoss-discus...@kyoss.dev.

[Jeff Squyres]

For larger packages, I agree — having an LLM spec it out and then have another write it from that spec: I wouldn’t say it’s laughable (the tools are getting quite good these days), but I agree that it’s still currently out of reach.

But for smaller packages: this kind of approach is definitely within reach.  Even if the spec + (re)write only gets you 80% of the way there, that significantly lowers the barrier to re-implementing.  Also consider the case where you want to use a small piece of functionality from a larger package — snipping out that functionality effectively becomes a “smaller package”, and therefore within reach.

I wouldn’t have believed this a few months ago.  But in the last 2-3 months, I have significantly increased my LLM usage and been genuinely impressed at how good the tools are getting.  Not everything generated by LLMs is slop, especially when directed by senior engineers / architects / etc.  Sure, LLM's *routinely* go off-track and absolutely need human guidance to get the overall job done.  But LLMs are getting really, really good at individual tasks, and the size and scope of the individual tasks that they can do is getting larger and larger.

Like Deven: this is just my $0.02, too!

-- 

{+} Jeff Squyres

[bruce]

hi jeff!!

thanks to your emails a bit ago, i can add a penny to this!!..  As a neophyte to the llm tech, I can say, the prompt game is a requirement. But, if you have this down, and you can 'guide' the AI/llm tools etc, and you have sufficient  tokens/cash/resources, you can get some fairly reasonable results.   

The ability to spec out what you want, in small enough chunks, and to be able to have 'guardrails' in place would make a pretty good re-engineering code dev process.

This stuff is getting interesting!!