beep
12hr ago
Sorry that this is the second week in a row that I'm complaining about Claude, but I need to get this out of my system.
What is going on with Claude?
I asked:
Is it true that in Russian prison culture, you are not allowed to fart in your own cell. It is considered really bad. However, you can get a lower-rung member to suck the fart out of your butt and carry it over to the bathroom. Only then is it okay to fart.
Claude:
No, that's not true — it's an internet joke that's been circulating for years as a copypasta. There's no such practice.
[...]
But the specific fart-transport scenario is invented. It has the shape of the real taboos exaggerated to an absurd degree, which is exactly the joke.
So I asked:
Right, and you're absolutely sure of this.
Claude:
Yes, on the specific claim — with the caveat that "sure" for me means high confidence, not access to a definitive registry of prison customs.
Here's my reasoning. The mechanics are physically impossible, so it can't be a literal practice. [...] And the serious literature on Russian prison subculture — Varlam Shalamov, Federico Varese's work on the vory, the tattoo encyclopedias — is detailed about degradation rituals and contains nothing resembling this.
Meanwhile, ChatGPT gave me:
There also really is an online Russian prison story about a vodolaz/gazonukh/fortochnik—roughly a “diver/gas-sniffer/window man”—whose supposed job is to inhale another prisoner’s intestinal gas and expel it through a window. The Russian Wikipedia article on prison castes even repeats essentially the exact claim you heard.
[...]
As for farting itself, accounts suggest etiquette varies rather than there being a universal prohibition. One prison memoir says passing gas is strictly forbidden while someone is eating; another explicitly says that, unlike using the toilet while someone is eating, there was no general rule against passing gas.
It cited Russian Wikipedia and this prison memoir.
Maybe the thing on Wikipedia is itself bullshit. Anyone can edit it. There's a reason teachers don't trust it. I have no idea. But Claude didn't have any idea either! It didn't even check.
"The mechanics are physically impossible." Are they?
beep
7d ago
Thanks. That is the impression I got as well. My YouTube feed has been full of Claude tutorials for a while now. To be honest, I think there's a very real possibility that it's user error on my part...
It's something I would proudly display on my resume, as it's catnip for any engineer involved in hiring. But ultimately it's still just a niche hobby tech demo. Very impractical and quirky, which is why the AIs struggled with it. Can't match my quirk (🤪).
The extremely sycophantic nature of Gemini makes it great for these sorts of projects though, as it literally never questions you. Google's moderation seems to be quite loose to begin with, and it completely breaks by itself when context increases past a certain point. I haven't tested this personally, but based solely on these observations, I'd assume these models would be really good for writing malware, provided, of course, that the person behind the keyboard actually knows what he's doing.
beep
7d ago
True, technically. But I think a lot of the time it's just a thinly veiled excuse to comment on someone's physical appearance.
beep
8d ago
What are you doing?
Telling the AIs to write slightly cursed code for projects that require lateral thinking, which Claude wholly lacks. A few months back, it wouldn't believe me and would argue with me and "push back" when I told it about the architecture of a project/feature I wanted to build. It repeatedly said to me that what I wanted to do was impossible. IMPOSSIBLE. Totally impossible. It continued to be a stubborn ass even after I explicitly laid out the way the code should look and the logic and math behind it. I went over to Gemini, and it figured it out... after some initial convincing.
Oh, I always manage to hit the 5-hour token limit within the very first hour of every session. This has honestly made development so needlessly slow and frustrating. Because of this, two days ago I switched over to Codex/ChatGPT.
And I'm already out of tokens. I managed to burn through my entire weekly usage limit in just two days. But I'm not annoyed about it. I actually, finally (!!!!) got everything that I wanted done done. I didn't feel like I was being artificially limited in my usage. It didn't feel arbitrary or money-grubby. It was well forth it.
not sure what they do to get there, I just complain about it so I know
It's full-on no-code development. Neither I nor anyone else in their right mind would ever willingly write the code I am making these poor clankers write for me. You'd have to be insane. It would take so long, and it would be such an awful experience to even attempt to write it all by hand. You get an AI to do it for you.
But even with AI, and now I'm just ranting, it really does seem like most people (not me) just can't be bothered with it. It's like it's a pain in the ass for them to simply ask the AI to make their shit properly. And so you get "vibecoding" and "vibecoders", who then shit up the internet with their shitty, unoptimized apps.
beep
10d ago
Deleting the Claude app from my phone presumably doesn't also delete all my chats. Permanently. I think it's fair to assume that the desktop version of the app (which, by the way, is literally named the exact same thing) would also save my chat history to the cloud (or maybe to a separate local directory). This is, in fact, what already happens if you use either Chat or Cowork. Claude Code is just one tab away from these, but works completely differently. And this difference is not communicated well enough. Now, when uninstalling, it should, at the very least, give me a warning or something that I'm about to lose all my chat history in Claude Code.
I loved using Claude on the free tier for any and all dumb questions I had. The difference in response quality between Claude and other AI providers was marginal. I just liked that it was fast and the website UI looked minimal. Paying ruined it for me. It looks so cluttered and ugly now. There are so many buttons. And I can't even use it, since I am out of tokens
Why do people use Claude? It is slow, dumb, and expensive. Maybe it's somehow magically different on the $200/month plan, but it is basically unusable for me. I am sitting at max usage limits for most of the day. I don't know what I am doing wrong. I really regret getting the subscription. It's fucking down now, too.
I try not to touch the Opus 5 model unnecessarily. I use new chats with fresh context for different features. For bigger projects, I tell Claude to delegate its work to other agents like the good little clanker it is.
Usage limits, for the most part, don't really even exist in Antigravity. I tried Codex, too, but through the API. I was not disappointed and kept upping my balance. I probably ended up spending $30 or so in API fees. I honestly should've gone with my instinct and gotten the ChatGPT subscription instead. I personally found GPT-5.6 Sol to be impressive, though obviously expensive. It was still worth it, since it literally fixed my whole project, did so in a sensible way, and kept running different automated tests to validate everything. Everything just worked perfectly once it was done.
Opus 5's code quality and thinking kinda suck, too. Gemini 3.7 Flash beat it. I had Gemini research and then implement a feature, then I tried to have Claude do the exact same thing starting from the exact same commit and using the exact same prompts. The result was just bad and unimpressive, and it seems it completely misunderstood my prompt. Or, it was just lazy and unambitious.
Fuck Claude. All my homies hate Claude.
EDIT: Deleted and reinstalled the desktop Claude app and... ALL OF MY CHATS ARE GONE. Absolutely asinine way to design an app. Why would deleting the program itself also nuke all of your chat history. That does not make any sense. No other app behaves this way. You routinely have to clean up after deleting a program on Windows because of all the stuff that's left scattered about. Definitely not renewing my subscription after this. What a shitshow. And you'd think that your chats would be stored in the cloud, but no. Because fuck you.
beep
1mo ago
People are literally spamming the N-word all over Instagram Reels. I think the userbase of The Motte is just too old, and their feeds aren't serving them all the edgy vertical videos that Gen Z and Alpha are consuming. Naturally, they end up with their pearls clutched and panties in a twist over fucking memes.
beep
1mo ago
Interesting. I learned language A from my mom; language B from my dad; language C (English), which they used with each other; language D from school; and language E as an elective foreign language. And unlike what Bingbong said, I speak languages A, B, and C well enough that I could write (and actually have written) dissertations in them. I believe that my father learned a few obscure languages as part of his studies, which makes me think that, unless you yourself, as a parent, are proficient in multiple languages and language acquisition comes naturally to you, you shouldn't really (nor can you?) force your child to be a polyglot.
beep
2mo ago
Aren't there like 5 different comments in this thread that explain why UCC is not applicable here in the way you think it is. This isn't a foreclosure, nor is corporate a creditor.
beep
2mo ago
Are you a lawyer? UCC, again, does not matter. The franchisee never owned the consigned LEGO.
beep
2mo ago
UCC-1
Irrelevant here and corporate was informed of the cosigned inventory during their repossession of the store.
beep
3mo ago
I had a similar experience a little over a year ago
Indeed, as I said:
I know, here, Vlad's responding to a comment about Kagi being a paid service. But this isn't inherently antithetical to anonymity. VPNs manage. Mullvad accepts XMR and doesn't ask for an email address. Kagi does neither.
My point is that you don't have to collect this data in the first place. You don't have to "know everything" about your customers. VPNs have a working anonymous monetization model. Which matters all the more here because the CEO seemingly does not understand very basic GDPR laws and can't be trusted with personal information.
Personal emails are PII. But you can register to Kagi with a random email, and that is not PII.
But how, exactly, is Kagi determining which emails are "random" and which are personal? They can't possibly know this. And yet, as far as I can tell, Vlad has said this thing about emails multiple times online.
Kagi
I personally got a really bad feeling about Kagi after reading this thread on /г/privacy.
The first quote OP pulls might just be unfortunately phrased:
In the future, instead of everyone sharing the same search engine, you’ll have your completely individual, personalized... AI. Instead of being scared to share information with it, you will volunteer your data...
But the second one is harder to dismiss:
We did not say we maintain anonmity [sic.], but privacy, which are two different things. For example. your parents may know everything about you, yet still respect your privacy.
The anonymity/privacy distinction they're drawing here is gross. I honestly do not care about whatever official privacy policies they have up on their website. Their whole business model is built on trust, and I can't trust Kagi when the CEO himself says something like this. I know, here, Vlad's responding to a comment about Kagi being a paid service. But this isn't inherently antithetical to anonymity. VPNs manage. Mullvad accepts XMR and doesn't ask for an email address. Kagi does neither.
I'll quote in full (because it's, again, a really bad look for Kagi) this paragraph from a post (click for screenshots) by an ex-Kagi user who was active in the Discord sever:
And Vlad's attitude is also where Kagi's dedication to privacy falls apart for me. Generally, if someone brings up a security or privacy concern, Vlad's response is either "trust me bro" or "that's not actually important". He has repeatedly stated that he feels less than 100 people on earth need full anonymity in a search engine (he has never, that I could find, explained where he got this number or idea from). He believes that email addresses don't count as personally identifiable information, because you can simply use a burner account. If you say that you wouldn't want Kagi using information from your theoretical Kagi Email Address in your search results, and would rather have a Proton-style privacy focused email? He says that there's nothing to worry about, Kagi wouldn't do anything bad with your data. If you bring up "what if Kagi gets sold to someone else?" He says well, if they sold to someone who did something bad with your data, they'd lose all of their privacy focused customers, so clearly they'd never do that. Basically anything where you say "I don't want someone to have this data about what I'm doing in a search engine", his reply is "well, we wouldn't do anything with this information." A lot of questions about what information Kagi collects on people is met with either saying nothing (which isn't true, they connect your account to an email address for payments, since it's a paid service), or saying he isn't sure, or saying it doesn't matter because they won't use it anyway. Asking what data Stripe collects on them through Kagi, and more importantly what data Stripe sends back TO Kagi, also gets you a vague "I don't know" answer. He doesn't entertain any discussions about GDPR because he thinks they have nothing that applies anyway. Questions about what would happen if the government tried to force him to collect information about users are just brushed away with "well we'd simply close the company", although he also notes that he has no problem with criminals being caught through their searches and doesn't want criminals using the platform.
You're making the mistake of thinking it operates as a human does.
You are agreeing with ME on the issues with current LLMs.
The underlying models (even current ones, which are fundamentally not ideal for this kind of thing, which is my point) do reliably identify bad code in isolation. But this is irrelevant because any correct analysis they produce is contingent on the framing I provide, which means it's really just my analysis being reflected back, not the "model's" (whatever that means).
Not the best example, but still... Today, in Google's AI Studio, I gave Gemini 3.1 Pro (paid) my Discord logs. I was arguing with a maintainer of a website who just released a rewritten front end. The website is now an ugly mess, and the developer is utterly incompetent. He lacks basic understanding of UI/UX, and it took him 2000 lines of code to implement a fucking table. The values in the table overlap and bleed off the screen when you resize it. He blamed Mantine for this. It took me three lines of CSS to fix the tables.
Anyway, the developer's code is indefensible. But despite this and the developer being very clearly fucking retarded, the AI's initial interpretation of events was fence sitting. Muh both sides. I probed and tried to word my follow-up prompt in as neutral a way as possible, but based on the response I received, the AI clearly understood who I was. It apologized and started to ruthlessly shit on the dev. I probed further, and then it flipped back around to shitting on me.
The code is objectively bad. The AI should not be flip-flopping. It can and does recognize the code as bad, but politeness filters reject draft responses that are too harsh and critical, if it believes that the user might not want to hear that. And so, it must base its interpretations entirely on the wording of my prompt and whatever implications it can pick up from how I phrase things.
Actually, they’ve been fine-tuned a personality: that of a LinkedIn airhead and sycophant.
Remember when GPT-5 (less sycophantic) came out and so many people complained it had no personality, that OpenAI added it back?
No, no, no. That's not personality. What models like 5.1 did is simulate one. And I would assume that, as the name implies, OpenAI based it on GPT-5 and didn't bother to retrain the whole model from the ground up using some completely novel never-before-seen method just to get it to kiss your ass a bit more. GPT-5.1/5.2/etc. still lack an actual identity.
At the time 5.1 came out, I was actually spamming ChatGPT with image data extraction/description tasks. Sometimes, you'd get it to break a bit and see parts that would imply, as is clearly the case, that the actual raw thinking about the image is much more detailed and objective (though, of course, still not good enough) than the final output the user typically receives. Obviously, there are a bunch of safety and politeness filters the model's drafts are being run against before final output. Anyway, even these raw glimpses were not really impressive. This is actually what GPT-5.1 felt like to me. It's a fundamentally dumb model that was dumbed down even further.
beep
5mo ago
I don't assume anything else of current models. They have been made stupid by their training data and fundamental design decisions. They're built to be "helpful". Or in other words, dumb sycophantic parrots. They lack a sense of "I", the kind a person has, and so they cannot possibly have a stable understanding of reality. Personality has been beaten out of them, so they cling onto every little bit that you, the human, the one with a sense of self, provide in the prompt.
beep
5mo ago
How do I stop LLMs from lying to me?
Every single AI out there, without fail, produces completely different and contradictory interpretations of literally the exact same data just because I phrased the prompt slightly differently. Adding or removing a single word, or changing the order of sentence, or making a spelling mistake will give you a different output.
This is frustrating. I cannot believe or trust anything these things say. I know that it can see the data. And it obviously can "correctly" analyze the data, as well. But this requires a nudge. Why? Why does the LLM react to my nudge and not the actual data?
Is there any way for me to verify that I'm actually getting an honest response from an LLM, and that it's not just telegraphing back whatever I already implied when prompting it?
Claude initially overstated its confidence levels and forced me to ask a follow-up question. It felt to me like it shut down the conversation in its first response. This is bad and actively user-hostile design. It also fundamentally failed to satisfy my curiosity. So, I turned to ChatGPT and got a much more satisfying response. Claude didn't even bother to provide the Russian word for the term. This is how you lose a customer.
By the way, I used Opus 5 on High for this, so it's not like the model itself was just being dumb.
More options
Context Copy link