Belle Belledonne ce soir !

There is some discussion around what this statement by a PHP developer means in practice:

“while a substantial amount of AI has been used for analysis, identification of vulnerable code, patch verification, and even test generation, only a small fraction of the functional changes were AI-generated and all of it was thoroughly reviewed and tested.”

https://github.com/php/php-src/commit/31c0ea7d7ff1d1fccb6492e52dab544011628065#commitcomment-206067056

I only skimmed this massive patch but other people were more thorough and found it… let’s say confusing. Changes that seem to be completely unrelated. Changes that seem to make no sense. Changes that seem inconsistent and/or incomplete. Complete lack of useful documentation. In other words: telltale signs of slop.

So now the stance is that only a small fraction of these changes is actually slop, and that this slop has been thoroughly reviewed and matches the quality requirements. Which, if true, would mean that the PHP codebase currently is indistinguishable from slop, with the code being incomprehensible to humans an expected occurrence. Mind you, I’m not ruling out this possibility, but it certainly is a concerning one.

Unfortunately, I consider it more likely that the PHP team is still understating the amount of LLM involvement in the creation of this patch. Which may or may not have been avoidable given the timeline and the state of the PHP codebase. If it is deemed unavoidable, then this too is very bad news for the PHP project: we are bound to see more large patches meant to eliminate the lamented technical debt. And since we’ve already established that the task is too complicated for humans you can guess who is going to be writing those.

Either way, this is the point where PHP as a project is making a decision pro or contra slop. They seem to be leaning towards the former, and once they go there it will be almost impossible to get back.

This thread summerised for normies

There’s a high-profile legal case in the UK High Court at the moment with a celebrity suing to prevent publication of allegations against them or their name.. but Google AI just apparently outright names them and the details when you click the headlines.

$30k for a new Mac Studio

Don’t worry in two years time it’ll probably be $100k

Elon Musk tried getting race hate figure Tommy Robinson out of jail by funding his defence, and lost again. https://www.bbc.co.uk/news/articles/cq2d6z36pp53o

🔎

#MakroMittwoch #photography #MacroPhotography #naturephotography #BlackAndWhitePhotography

RE: https://cyberplace.social/@GossiTheDog/117514165293840633

Anthropic made:

- 1.5 billion USD losses in 2023
- 8.3 billion USD losses in 2024
- 42.0 billion USD losses in 2025
- 80.7 billion USD losses in the first half of 2026 alone
- losses in every year since its founding

Anthropic roughly quadruples its losses every year. I'm not saying that this trend will continue or that Anthropic will make 640 billion USD losses in 2027.

Just that per its own admission, Anthropic may never be profitable.

But "the genie is out of the bottle, it can't be put back into the bottle." "AI is here to stay." It's very intelligent of business leaders to fire the only people who still understand their codebase and have the institutional knowledge of how their employer's business works. And let the others continue fuck up their codebase. No, no. It won't be an expensive, maybe even existential problem at all, once the business leaders can no longer afford the true price of unsubsidized LLMs, to work with and unfuck a codebase literally nobody in their company understands anymore and everyone is afraid to touch/change in any way, or to re-gain the lost institutional knowledge. No, no.