I think LLMs can help with keeping up to date with spec changes, but that is not something that is being explored. Instead we apparently do line-by-line ports of existing codebases, which we already know LLMs are good at.

3/???

html5lib is already 100% Python, but (like html5ever) is lagging behind on the spec. I think the reason is that updating parsers to conform with the latest spec changes is a very unrewarding task, and it's hard to get back into it after maybe a few months of working on it.

2/2 ...

ah damnit i was hoping for some decent web-related benchmarking input

[...] much harder to figure out programmatically if a given seed helped the fuzzer, and that the fuzzer didn't find the same input before the LLM did.

I think the biggest challenge I have is extracting useful signal from the fuzzer to guide the LLM. Like, the LLM can of course find new seeds and validate by itself that they crash the target, but it's [...]

also what is that vercel dataset you're referring to, is it public? the linked page doesn't seem to be very informative.

that really depends on who you think really wants to cure cancer here and who is acting purely out of monetary incentive. like, do you think those companies want to replace workers because it will lead us to a cure faster?

are these results consistent with Linux?

3

lol gotta secure that curl|sh

you'd think that if apple hooked into executables already, and even apparently modified their versions of tar and unzip to propagate the quarantine bit, they'd patch the builtin shell to check the quarantine bit too when executing arbitrary code. or any of the other interpreters they ship...

lobste.rs

You wouldn't believe the shit I get send each week from people working in industry/gov. We are entirely unprepared for the mess this is creating.

I'd send them to journalistic outlets with their inside info, but they too are entirely incapable of analyzing and presenting all of this properly.