@timbray In #curl we see good in code reviews by LLMs. But they generate a flood initially. There are issues they can find and others they are blind to
New is that they find discrepancies between comments and code. In something as old and alive as #curl this is very useful. No other tool can do that.
But we get access for free, having a brand. These are not cheap. And we could not pay for them, so…
AI economics are nuts. Bananas.
@timbray instead of focusing on LLM-driven development, I'd like to see more people talking very loudly about what scraping-for-training-LLMs is doing to open source infrastructure. We've had to block over a million IP addresses and bar access to individual commits for all our public repos. Our bandwidth costs are driven up. Every FLOSS project I know is facing this. They don't respect our licensing, our time or our money. This is a problem *right now*, not next month or next year.
@PaulDavisTheFirst Absolutely, but it's not just an OSS problem, it's a freaking Internet problem. News sites, public services in healthcare and so on, the Internet Archive. Obviously a certain amount of legislation/litigation would help. But my biggest hope is that when the bubble pops, all this costly crawling will drop by an order of magnitude.
@timbray yeah there’s room, but very little incentive to build one. Like Linux or KDE or Wikipedia, we might persist and do it anyway, but I’m not holding my breath.
2/2 … and in which I describe the second of the two Quamina-related Claude interventions, namely an automated port from Go to Rust: https://www.tbray.org/ongoing/When/202x/2026/02/14/Q-Plus-C-Ch2
@timbray What's the copyright on those translations?
@timbray I’m glad you published this, bc it’s a very weird moment where one minute it feels like the bots empower me and the next like they’re coming after me. On balance, i think it’s empowering, like any knowledge sharing. But there is an implosion taking place in sw development.
(Now I’m in trouble, probably.)
@timbray In #curl we see good in code reviews by LLMs. But they generate a flood initially. There are issues they can find and others they are blind to
New is that they find discrepancies between comments and code. In something as old and alive as #curl this is very useful. No other tool can do that.
But we get access for free, having a brand. These are not cheap. And we could not pay for them, so…
AI economics are nuts. Bananas.
@icing AI technology may or may not make sense but I totally can't see sense in the economics.
