If your thing ends up actually good you now have a defence against exploitation, and a way to generate income reliably (by selling the code under a different license). afaik, organisations like the FSF even endorse this.

This has always been the case. Sometimes they give back by opening one or more of their components. Other times they don't. I don't see it as a problem. It doesn't usually detract from what's already published.

In cases where it would detract, simply use an appropriate license to curb the behavior.

> LLMs are just accelerating this trend.

LLMs might not prove sufficiently capable to meaningfully impact this dynamic.

Alternatively, if they achieve that level then I think they will accomplish the long stated goal of FOSS by enabling anyone to translate constraints from natural language into code. If I could simply list off behaviors of existing software and get a reliable reproduction I think that would largely obsolete worrying about software licenses.

I realize we're nowhere near that point yet, and also that reality is more complex than I'm accounting for there. But my point is that I figure either LLMs disrupt the status quo and we see benefits from it or alternatively that business as usual continues with some shiny new tools.

As incentivized by temporarily-free licenses.