Stubsack: weekly thread for sneers not worth an entire post, week ending 22nd February 2026
-
Scott Shambaugh mulls about an AI alignment issue following his run-in with a bot last week
-
i’ve collided with an article* https://harshanu.space/en/tech/ccc-vs-gcc/
you might be wondering why it doesn’t highlight that it fails to compile linux kernel, or why it states that using pieces of gcc where vibecc fails is “fair”, or why it neglects to say that failing linker means it’s not useful in any way, or why just relying on “no errors” isn’t enough when it’s already known that vibecc will happily eat invalid c. it’s explained by:
Disclaimer
Part of this work was assisted by AI. The Python scripts used to generate benchmark results and graphs were written with AI assistance. The benchmark design, test execution, analysis and writing were done by a human with AI helping where needed.
even with all this slant, by their own vibecoded benchmark, vibecc is still complete dogshit with sqlite compiled with it being slower up to 150000x times in some cases
This is why CCC being able to compile real C code at all is noteworthy. But it also explains why the output quality is far from what GCC produces. Building a compiler that parses C correctly is one thing. Building one that produces fast and efficient machine code is a completely different challenge.
Every single one of these failures is waved away because supposedly it’s impressive that the AI can do this at all. Do they not realize the obvious problem with this argument? The AI has been trained on all the source code that Anthropic could get their grubby hands on! This includes GCC and clang and everything remotely resembling a C compiler! If I took every C compiler in existence, shoved them in a blender, and spent $20k on electricity blending them until the resulting slurry passed my test cases, should I be surprised or impressed that I got a shitty C compiler? If an actual person wrote this code, they would be justifiably mocked (or they’re a student trying to learn by doing, and LLMs do not learn by doing). But AI gets a free pass because it’s impressive that the slop can come in larger quantities now, I guess. These Models Will Improve. These Issues Will Get Fixed.
-
and people keep telling me this shit is good
I mean, this one is really good, I got like half an hour of jokes with my friend off it
okay I can’t argue with that outcome
-
AI bros do new experiments in making themselves even stupider. Going from ‘explain what you did but dumb it down for me and my degraded attention span’ into ‘just make a simplified cartoon out of it’.
Proud of not understanding what is going on. None of these people could hack the Gibson.
E: If they all hate programming so much, perhaps a change of job is in question, sure might not pay as much, but it might make them happier.
my current favorite trick for reducing “cognitive debt” (h/t @simonw ) is to ask the LLM to write two versions of the plan:
- The version for it (highly technical and detailed)
- The version for me (an entertaining essay designed to build my intuition)
I don’t know about them, but I would be offended if I was planning something with a collaborator, and they decide to give me a dumbed down, entertaining, children’s storybook version of their plan while keeping all the technical details to themselves.
Also, this is absolutely not what “cognitive debt” means. I’ve heard technical debt refers to bad design decisions in software where one does something cheap and easy now but has to constantly deal with the maintenance headaches afterwards. But the very concept of working through technical details? That’s what we call “thinking”. These people want to avoid the burden of thinking.
-
my current favorite trick for reducing “cognitive debt” (h/t @simonw ) is to ask the LLM to write two versions of the plan:
- The version for it (highly technical and detailed)
- The version for me (an entertaining essay designed to build my intuition)
I don’t know about them, but I would be offended if I was planning something with a collaborator, and they decide to give me a dumbed down, entertaining, children’s storybook version of their plan while keeping all the technical details to themselves.
Also, this is absolutely not what “cognitive debt” means. I’ve heard technical debt refers to bad design decisions in software where one does something cheap and easy now but has to constantly deal with the maintenance headaches afterwards. But the very concept of working through technical details? That’s what we call “thinking”. These people want to avoid the burden of thinking.
Eh, one might say that going by the broad strokes version while letting the expert do their thing is basically what management is all about, especially if they ignore the part where he wants his version to be light and entertaining.
This isn’t about managing subordinates though, this is about devising ways to be complacent about not double checking what the LLM generates in your name.
-
This is why CCC being able to compile real C code at all is noteworthy. But it also explains why the output quality is far from what GCC produces. Building a compiler that parses C correctly is one thing. Building one that produces fast and efficient machine code is a completely different challenge.
Every single one of these failures is waved away because supposedly it’s impressive that the AI can do this at all. Do they not realize the obvious problem with this argument? The AI has been trained on all the source code that Anthropic could get their grubby hands on! This includes GCC and clang and everything remotely resembling a C compiler! If I took every C compiler in existence, shoved them in a blender, and spent $20k on electricity blending them until the resulting slurry passed my test cases, should I be surprised or impressed that I got a shitty C compiler? If an actual person wrote this code, they would be justifiably mocked (or they’re a student trying to learn by doing, and LLMs do not learn by doing). But AI gets a free pass because it’s impressive that the slop can come in larger quantities now, I guess. These Models Will Improve. These Issues Will Get Fixed.
spent $20k on electricity blending them
They would probably be even more impressed that you only spent $20k
-
in today’s news about magical prompts that super totes give you superpowers:
We introduced SKILLSBENCH, the first benchmark to systematically evaluate Agent Skills as first-class artifacts. Across 84 tasks, 7 agent-model configurations, and 7,308 trajectories under three conditions (no Skills, curated Skills, self-generated Skills), our evaluation yields four key findings: (1) curated Skills provide substantial but variable benefit (+16.2 percentage points average, with high variance across domains and configurations); (2) self-generated Skills provide negligible or negative benefit (–1.3pp average), demonstrating that effective Skills require human-curated domain expertise
I am jack’s surprised face
…and given I have other yaks, I shall not step on my “software and tools don’t have to suck” soapbox right now
This reminds me of when Steve Jobs would introduce every new Mac release by talking about how fast it could render in Photoshop. I wonder how he would do in our brave new era of completely ass-pulling your own bespoke benchmark frameworks.
-
who up continvoucly morging they branches
That slopped-out “diagram” plagiarised Vincent Driessen’s “A successful Git branching model”, BTW.
-
AI bros do new experiments in making themselves even stupider. Going from ‘explain what you did but dumb it down for me and my degraded attention span’ into ‘just make a simplified cartoon out of it’.
Proud of not understanding what is going on. None of these people could hack the Gibson.
E: If they all hate programming so much, perhaps a change of job is in question, sure might not pay as much, but it might make them happier.
E: If they all hate programming so much, perhaps a change of job is in question, sure might not pay as much, but it might make them happier.
Surely at least a few of them have worked up enough seed capital to try their hand at used-car dealerships. I can attest that the juicier markets just outside the Bay Area are fairly saturated, but maybe they could push into lesser-served locales like Lost Hills or Weaverville.
-
A little exchange on the EA forums I thought was notable: https://forum.effectivealtruism.org/posts/EDBQPT65XJsgszwmL/long-term-risks-from-ideological-fanaticism?commentId=b5pZi5JjoMixQtRgh
tldr; a super long essay lumping together Nazism, Communism and religious fundamentalism (I didn’t read it, just the comments). The comment I linked notes how liberal democracies have also killed a huge number of people (in the commenter’s home country, in the name of purging communism):
The United States presented liberal democracy as a universal emancipatory framework while materially supporting anti-communist purges in my country during what is often called the “Jakarta Method". Between 500,000 and 1 million people were killed in 1965–66, with encouragement and intelligence support from Western powers. Variations of this model were later replicated in parts of Latin America.
The OP’s response is to try to explain how that wasn’t real “liberal democracy” and to try to reframe the discussion. Another commenter is even more direct, they complain half the sources listed are Marxist.
A bit bold to unqualifiedly recommend a list of thinkers of which ~half were Marxists, on the topic of ideological fanaticism causing great harms.
I think it’s a bit bold of this commenter to ignore the empirical facts cited in how many people ‘liberal democracies’ had killed and to exclude sources simply for challenging your ideology.
Just another reminder of how the EA movement is full of right wing thinking and how most of it hasn’t considered even the most basic of leftist thought.
Just another reminder of how the EA movement is full of right wing thinking and how most of it hasn’t considered even the most basic of leftist thought.
I continue to maintain that EA boils down to high-dollar consumerism focused on intangible goods. I’m sure that statement won’t fly on LW or any other EA forum, but my thoughts on psychiatry don’t fly at a Scientologist convention either.
-
The Dangerous Economics of Walk-Away Wealth in the AI Talent War
Watch now | How firms are accidentally paying their best employees to become their biggest competitors.
(softcurrency.substack.com)
- Anthropic (Medium Risk) Until mid-February of 2026, Anthropic appeared to be happy, talent-retaining. When an AI Safety Leader publicly resigns with a dramatic letter stating “the world is in peril,” the facade of stability cracks. Anthropic is a delayed fuse, just earlier on the vesting curve than OpenAI. The equity is massive ($300B+ valuation) but largely illiquid. As soon as a liquidity event occurs, the safety researchers will have the capital to fund their own, even safer labs.
WTF is “even safer” ??? how bout we like just don’t create the torment nexus.
Wonder if the 50% attrition prediction comes to pass though…
the capital to fund their own, even safer labs.
I wonder, is this a theory of “safety” analogous to what’s driven the increased gigantism of vehicles in the US? Sure seems like it.
-
That slopped-out “diagram” plagiarised Vincent Driessen’s “A successful Git branching model”, BTW.
It’s funny that such a thing is rare enough in the corpus to come out so recognizably in the output.
-
This is why CCC being able to compile real C code at all is noteworthy. But it also explains why the output quality is far from what GCC produces. Building a compiler that parses C correctly is one thing. Building one that produces fast and efficient machine code is a completely different challenge.
Every single one of these failures is waved away because supposedly it’s impressive that the AI can do this at all. Do they not realize the obvious problem with this argument? The AI has been trained on all the source code that Anthropic could get their grubby hands on! This includes GCC and clang and everything remotely resembling a C compiler! If I took every C compiler in existence, shoved them in a blender, and spent $20k on electricity blending them until the resulting slurry passed my test cases, should I be surprised or impressed that I got a shitty C compiler? If an actual person wrote this code, they would be justifiably mocked (or they’re a student trying to learn by doing, and LLMs do not learn by doing). But AI gets a free pass because it’s impressive that the slop can come in larger quantities now, I guess. These Models Will Improve. These Issues Will Get Fixed.
Building a compiler that parses C correctly is one thing. Building one that produces fast and efficient machine code is a completely different challenge.
Ye, the former can be done in a month of non-full-time work by an undergrad who took Compilers 101 this semester or in literally a single day by a professional, and the latter is an actual useful product.
So of course AI will excel at doing the first one worse (vibecc doesn’t even reject invalid C) and at an insane resource cost.
-
Want to wade into the snowy surf of the abyss? Have a sneer percolating in your system but not enough time/energy to make a whole post about it? Go forth and be mid.
Welcome to the Stubsack, your first port of call for learning fresh Awful you’ll near-instantly regret.
Any awful.systems sub may be subsneered in this subthread, techtakes or no.
If your sneer seems higher quality than you thought, feel free to cut’n’paste it into its own post — there’s no quota for posting and the bar really isn’t that high.
The post Xitter web has spawned so many “esoteric” right wing freaks, but there’s no appropriate sneer-space for them. I’m talking redscare-ish, reality challenged “culture critics” who write about everything but understand nothing. I’m talking about reply-guys who make the same 6 tweets about the same 3 subjects. They’re inescapable at this point, yet I don’t see them mocked (as much as they should be)
Like, there was one dude a while back who insisted that women couldn’t be surgeons because they didn’t believe in the moon or in stars? I think each and every one of these guys is uniquely fucked up and if I can’t escape them, I would love to sneer at them.
(Credit and/or blame to David Gerard for starting this. Also, hope you had a wonderful Valentine’s Day!)
-
who up continvoucly morging they branches
-
what I’m thinking about is for how many years now they have been promising that just one more datacenter will fix the “hallucinations”, yet this mess is indistinguishable from nonsense output from three years ago. I see “AI” is going well
you can count on microslop to always be behind the curve
-
Want to wade into the snowy surf of the abyss? Have a sneer percolating in your system but not enough time/energy to make a whole post about it? Go forth and be mid.
Welcome to the Stubsack, your first port of call for learning fresh Awful you’ll near-instantly regret.
Any awful.systems sub may be subsneered in this subthread, techtakes or no.
If your sneer seems higher quality than you thought, feel free to cut’n’paste it into its own post — there’s no quota for posting and the bar really isn’t that high.
The post Xitter web has spawned so many “esoteric” right wing freaks, but there’s no appropriate sneer-space for them. I’m talking redscare-ish, reality challenged “culture critics” who write about everything but understand nothing. I’m talking about reply-guys who make the same 6 tweets about the same 3 subjects. They’re inescapable at this point, yet I don’t see them mocked (as much as they should be)
Like, there was one dude a while back who insisted that women couldn’t be surgeons because they didn’t believe in the moon or in stars? I think each and every one of these guys is uniquely fucked up and if I can’t escape them, I would love to sneer at them.
(Credit and/or blame to David Gerard for starting this. Also, hope you had a wonderful Valentine’s Day!)
A longread on AI greenwashing begins thusly:
The expansion of data centres - which is driven in large part by AI growth - is creating a shocking new demand for fossil fuels. The tech companies driving AI expansion try to downplay AI’s proven climate impacts by claiming that AI will eventually help solve climate change. Our analysis of these claims suggests that rather than relying on credible and substantiated data, these companies are writing themselves a blank cheque to pollute on the empty promise of future salvation. While the current negative effects of AI on the climate are clear, proven and growing, the promise of large-scale solutions is often based on wishful thinking, and almost always presented with scant evidence.
(Via.)
-
Want to wade into the snowy surf of the abyss? Have a sneer percolating in your system but not enough time/energy to make a whole post about it? Go forth and be mid.
Welcome to the Stubsack, your first port of call for learning fresh Awful you’ll near-instantly regret.
Any awful.systems sub may be subsneered in this subthread, techtakes or no.
If your sneer seems higher quality than you thought, feel free to cut’n’paste it into its own post — there’s no quota for posting and the bar really isn’t that high.
The post Xitter web has spawned so many “esoteric” right wing freaks, but there’s no appropriate sneer-space for them. I’m talking redscare-ish, reality challenged “culture critics” who write about everything but understand nothing. I’m talking about reply-guys who make the same 6 tweets about the same 3 subjects. They’re inescapable at this point, yet I don’t see them mocked (as much as they should be)
Like, there was one dude a while back who insisted that women couldn’t be surgeons because they didn’t believe in the moon or in stars? I think each and every one of these guys is uniquely fucked up and if I can’t escape them, I would love to sneer at them.
(Credit and/or blame to David Gerard for starting this. Also, hope you had a wonderful Valentine’s Day!)
The phrase “ambient AI listening in our hospital” makes me hear the “Dies Irae” in my head.
-
can all of rationalism be reduced to logorrhea with load-bearing extreme handwaving (in this case, agentic self preservation arises through RL scaling)?
-
OT: I tried switching away from my instance to quokk.au because it becomes harder and harder to justify being on a pro-ai instance, but it seems that your posts stopped federating recently.
Which makes that a bit harder to do. Any clue why?bluemonday1984@awful.systems update: quokka reached out to me and apparently you had been banned on another instance for report abuse and that ban had synchronised to quokk.au. You should be unbanned now which means the next stubsack should federate again.
E: I do not know how tags work E2: why does that format to a mailto link?
![Jack Riddle[Any/All]](https://lemmy.dbzer0.com/pictrs/image/ca54ea94-6df3-4a0f-b75a-f11c4f6a8e32.webp "Jack Riddle[Any/All]"){kind=avatar}
