Posts by jonny@neuromatch.social
(DIR) Post #B4griV1ERtIgl7v2Vk by jonny@neuromatch.social
1 likes, 0 repeats
the thing about all these people who love LLMs is that they will shit in your mouth, ask you to consider the taste of the shit they shat in your mouth, spend your hard-won moments of human life to let the shit they shat in your mouth sink into your mouth, and then they will take whatever you say and somehow turn that into more shit in your mouth and pretend like you weren't just eating the shit they just shat in your mouth already
(DIR) Post #B4griXmSBIGHKA1mOO by jonny@neuromatch.social
1 likes, 0 repeats
i just want to remind all the AI maximalists out there that you guys are fucking assholes. like all of you are huge dickheads. like not as an immutable trait, just that using AI seems to make you a huge fucking asshole, and you all act like dickheads to everyone and don't seem to even notice. like you are riding on a magic carpet powered by a dogs shit and everyone around you is like "wow why is that guy riding on a magic dogshit carpet around everyone and getting dogshit all over them" and you are just like "ha ha ha i am flying fuck you all"
(DIR) Post #B4pBMrCipuSvXhvkTw by jonny@neuromatch.social
8 likes, 5 repeats
Claude code source "leaks" in a mapfilepeople immediately use the code laundering machines to code launder the code laundering frontend now many dubious open source-ish knockoffs in python and rust being derived directly from the sourceWhat's anthropic going to do, sue them? Insist in court that LLM recreating copyrighted code is a violation of copyright???
(DIR) Post #B4q0VrjlJsMXSpRCBE by jonny@neuromatch.social
0 likes, 1 repeats
OK i can't focus on work and keep looking at this repo. So after every "subagent" runs, claude code creates another "agent" to check on whether the first "agent" did the thing it was supposed to. I don't know about you but i smell a bit of a problem, if you can't trust whether one "agent" with a very big fancy model did something, how in the fuck are you supposed to trust another "agent" running on the smallest crappiest model? That's not the funny part, that's obvious and fundamental to the entire show here. HOWEVER RECALL the above JSON Schema Verification thing that is unconditionally added onto the end of every round of LLM calls. the mechanism for adding that hook is... JUST FUCKING ASKING THE MODEL TO CALL THAT TOOL. second pic is registering a hook s.t. "after some stop state happens, if there isn't a message indicating that we have successfully called the JSON validation thing, prompt the model saying "you must call the json validation thing"this shit sucks so bad they can't even CALL THEIR OWN CODE FROM INSIDE THEIR OWN CODE. Look at the comment on pic 3 - "e.g. agent finished without calling structured output tool" - that's common enough that they have a whole goddamn error category for it, and the way it's handled is by just pretending the job was cancelled and nothing happened.
(DIR) Post #B4waiecy8OQdrtPv1M by jonny@neuromatch.social
0 likes, 0 repeats
It was only a matter of time before some AI-addled guy saw me quoting fucked up parts of the Claude code source, thought it was my code, and say it was just a skill issue and I just need to {prompting superstition} but it finally happened.
(DIR) Post #B4yYIPn4rdMIjNQRTE by jonny@neuromatch.social
1 likes, 0 repeats
RE: https://neuromatch.social/@jonny/116324676116121930Part 2 of exploring The Claude Code Source Leak Exclusion Zone continues here.(the reply tree under the prior thread is getting expensive to render and the bottom no longer renders unless you're logged in lol)end of prior thread: https://neuromatch.social/@jonny/116345400731237947RT: https://neuromatch.social/users/jonny/statuses/116324676116121930
(DIR) Post #B56fTGRmCI5FsiGIjI by jonny@neuromatch.social
0 likes, 1 repeats
It's so cool that anthropic is setting up a double-sided protection racket where it will profit from the massive token burn of attackers and defenders with a tool specifically designed to generate exploits and their only observable mitigation is a clientside system prompt that sternly warns the LLM to be good and not do malwarehttps://red.anthropic.com/2026/mythos-preview/
(DIR) Post #B56s3g0TqKMd1pciqO by jonny@neuromatch.social
0 likes, 0 repeats
RE: https://hachyderm.io/@jedbrown/116373136965443776Very good thank you information conglomeratesRT: https://hachyderm.io/users/jedbrown/statuses/116373136965443776
(DIR) Post #B5LoK1cJmMIEEuGEXA by jonny@neuromatch.social
0 likes, 0 repeats
the real skill with no curriculum is not "how to use consumer AI slot machines to get high on stolen valor" but actually "how to defend your code against the roving horde of opportunistic PR agents"
(DIR) Post #B5LoK2OAuQZedKeS80 by jonny@neuromatch.social
1 likes, 0 repeats
i need to write a paper about how the LLM turned what should have been a 1ksloc package written in a few weeks of having fun thinking about an interesting problem into a 6 month slog trimming a 10ksloc explosion of slop back down to 1ksloc package in one of the more mind numbing grinds i've ever done.
(DIR) Post #B666CroSZHuMFepiBU by jonny@neuromatch.social
2 likes, 0 repeats
The thing we actually need to get off github is not for more people to tell people to get off github, its for some of you with the technical expertise and tech company salary to pay for or volunteer to do the work to finish forgejo's federation so it's actually possible to compete with the kind of network and reputation effects needed to remain employed that are the actual thing github provides.
(DIR) Post #B6Dk1rsPLbclXKi2Qi by jonny@neuromatch.social
0 likes, 0 repeats
RE: https://mastodon.social/@hauschke/116558334793108142So far, US courts have mainly backed claims from AI firms that the way LLMs use copyrighted material is ‘transformative’, which is one of the tests for what counts as fair use.me reading a paper is transformative, in the sense that i transform the work into me having read the work. the purpose and character of my reading a paper is for exclusively nonprofit educational purposesthe nature of the copyrighted work is a claim of fact on shared, material reality, which cannot be considered privately held or created by the publisherthe effect of my reading the paper on the market is none, for the market does not consist of consumers directly paying for the copyrighted work.RT: https://mastodon.social/users/hauschke/statuses/116558334793108142
(DIR) Post #B6M7Db3pXvDB4KIF5U by jonny@neuromatch.social
0 likes, 0 repeats
The amount of dogs that can ride the train on their own is awesome
(DIR) Post #B6PNqiBl9z1pD5YUAy by jonny@neuromatch.social
0 likes, 0 repeats
So my little motorcycle that has been sitting neglected for a few years had its entire gas tank evaporate - no signs of rust in the tank or anything (plenty of rust on its screws, and the paint needs to be touched up), just looks like the gasket around the fuel cap disintegrated and maybe one of the lines has cracked a bit. So do I just replace all the lines and flush it with fuel cleaner? or how do you recover an engine from "total evaporation and everything is probably just gum now"
(DIR) Post #B6PtvALhbEB2EuNDma by jonny@neuromatch.social
0 likes, 0 repeats
I regret to give anyone this information that didn't already have it, but "the left for AI" as a cluster of ideas both a) exists and b) is infopoison of the highest order
(DIR) Post #B6qlbCCpuWvNClrU6C by jonny@neuromatch.social
1 likes, 1 repeats
RE: https://hails.org/@hailey/116657391001259044all the criticism has been said, all the takes been had. the only metaphor i have been finding consistently useful for understanding what is happening with people and "AI" is addiction, and specifically gambling addiction.RT: https://hails.org/users/hailey/statuses/116657391001259044
(DIR) Post #B6qlbCbeQGm2RjTImG by jonny@neuromatch.social
2 likes, 0 repeats
So, look. One shot rewriting the whole test suite in another language is probably not great to do, but what happened here is so much worse than you are expecting. https://github.com/RsyncProject/rsync/pull/903/This does not "translate tests into pytest" or a unit testing framework, it writes its own testing framework where tests are whole python scripts that redefine basic test functions in every script. Surely there would be a single way to "run rsync and get the results" - nope, well, there is, but then every test file will randomly redefine its own _run_and_capture function. So like now rsync needs a test suite for its test suite.If instead of telling an LLM to "rewrite the tests in python" you just searched "python testing" you would find the pytest docs. And then you would find examples. And then you could write fixtures to deduplicate all the prior shell script setup and teardown stuff, and so on. But since it was just "rewrite the tests in python" its now worse than before, and the odds of the rewrite actually being a 100% faithful translation are close to 0.
(DIR) Post #B8qBkvF73Q8qlLFrTk by jonny@neuromatch.social
0 likes, 0 repeats
@ariadne that's also the only way i've ever gotten it to do plausible work for things where that is definable, to already have a handwritten set of tests that define completion, and i've seen people say that's how they work as an evolution of TDD. for something like a game this simple i usually wouldn't even write tests except the most obvious regression tests for "the game still works". thinking about how to test it would be almost as complex as the game itself. i also agree about this not being how i would want to work in general - this has been a joyless endeavor that feels more like addiction than problem solving. the only place it "improves" over just writing the damn thing myself is that it required a hell of a lot less cognitive effort and thinking, but that... is... the fun part...