> Cloudflare, along with a majority of the world's leading publishers and AI companies, is changing the default to block AI crawlers unless they pay creators for their content.
It really seems that they did _not_ change the default, since this feature is in private beta.
CI/CD, private repositories, providing hosting. These are the options used by similar companies.
But I like that they’re focussing on creating something useful before chasing revenue. Once they’ve got a single tool that provides a consistent dev experience for Python developers and it’s widely adopted they should be able to pursue monetisation easily.
I think that's a bit optimistic; that's the path every VC-funded tooling company tries to take, and it often doesn't end well; restrictive licenses, hostile forks, early deaths, etc. You need to have some kind of plan ahead of time.
AFAIK ptmalloc (on which glibc is based) was created decades ago and both multi-threaded application and multi CPU systems where rare back then (at least in the Linux world) so multi-threaded performance didn't matter. Some improvement in glibc were made since then but I don't think it's possible to significantly improve glibc malloc without rewriting it more or less fully. At that point it would make more sense to import some existing malloc implementation.
And we are speaking about trafeoff the default number of arenas in glibc malloc is 8 times CPU which is a terrible tradeoff - on many workloads it cause heap fragmentation and memory usage (RSS) many times higher than allocated memory size, that's why it is common to find advice to set MALLOC_ARENA_MAX to 1 or 2. But probably such high number of areans allows glib to look less bad on synthetic benchmakrs.
Jemalloc, tcmalloc, mimalloc all were created with focus on multi-threaded applications from the beginning and while they don't work better than glibc malloc for single threaded application they don't work worse for this use case either. Probably the main disadvantage of using je/tc/mi mallocs for a single threaded app is large code size.
I switched an agent from Sonnet V2 to o3-mini (default medium mode) and got strangely poor results: only calling 1 tool at a time despite being asked to call multiple, not actually doing any work, and reporting that it did things it didn't
Do you mean that you did the research, or your AI product did the research? How are you assessing whether a death was plausibly murder? So strange that the one example you find is of a car crash (one of the most common ways people die!) and no citation that it was found to be murder.
