Monologue: The Agony Of GPT-5

I love making this show for all of you. It's been kind of a pisser, but it's important we get on top of this goddamn subject. So check the episode notes, buy a challenge coin, read the newsletter. A's a premium version unrelated to this show. I'd love if you'd subscribe, and if you don't, I won't feel anything. Don't worry about it, but let's get to it. Last week open Ai launched GPT five, a new flagship model of some sort that's allegedly better at coding and writing, but in

reality is much more of the same. Another model is interterminately better at benchmarks built specifically for large language moddles, because they can't do actual work. The Wall Street Journal reported late last year that it took multiple half billion dollar training runs to get GPT five off the ground, and Altman him said said in the podcast with theovon of all people at GPT five scared him and made him say, what have we done? And that's a good

bloody question, Sammy. According to open Ai, GPT five is a unified system with a smart, efficient model that answers most questions, a deeper reasoning model, and a real time router that quickly decides which model to use based on conversation type, complexity, tool needs. And you're explicit in tent. I read all of that out because I wanted you to hear how convoluted GPT five is and how much effort open Ai has had to put in to create something that, based on all reports is fine. To quote

Simon Willison, it just does stuff. Wowie zowie. In simpler terms, Chad GPT's version of GPT five takes the user's prompt and decides which modele to use as a result, using one of a few submodels GPT five Regular, Mini or now I know, and then spits out an output. And there are rate limits, by the way, so if you use it too much you get kicked down to many automatically. If you ask it to think about something, it will choose to engage the reasoning part of the model. These

things do not think, by the way. They are probabilistic models, so reasoning is kind of like you get a prompt and then it reads the prompt with another model and says, Okay, what would the steps be to execute this? It has some returns, but recent papers suggest that it doesn't work that well anyway. Using this, you mostly have to trust the open ai will choose the best model for the job, as opposed to the cheapest one for open ai to serve,

which is what I think they're actually doing. As part of the launch, open ai has also killed access to all other models that at least is planning to, and truncated user access to two choices GPT five or GPT five thinking with legacy models like four four many and their associated rate limits gone immediately for most, though some of them are back, and in sixty days of people paying two hundred dollars a month, you'll work out why

I'm dithering in a second. This enraged the chat gp subreddit, with users claiming that GPT five was and I quote, wearing the skin of their dead friend, referring to GPT four oh, and another saying that GPT four point five genuinely talk to them and as pathetic as it sounds, was their only friend. And it must be clear, we can make fun of these people if we want, but this is actually genuinely sad. There is something going on here where people are so lonely that they want to

talk to a chat bot. Mock them if you want, and some of you will, and I don't know if I even want to. But something is happening here and it isn't brilliant. But after a few days, Clammy Samuel and restored access to GPT four overpaid users, and then this only matched the stem the type briefly, with one user saying that their baby was back, that they cried a lot, and that they were crying as they wrote the post, ending by saying love you. I assumed to

GPT four zero. Here's the problem. Though, users are now doubting that the four O that open Ai is restored is actually the same model. One post claims that four roh has lost its soul, and another says that fourro is lobotomized all the way down in one three. One user said that four row had gotten markedly worse suddenly, and another said it's definitely not the same, though others

in the thread claimed that it was. Another post said legacy GPT four row is GPT five in cosplay, even as others pleaded with them and said it was exactly the same and that people were experiencing some strange phantom love for placebo effect. And I think that's actually what's

going on, writ large. Chat GPT was never a success based on its actual abilities or outputs or things it could do, but a global marketing campaign perpetuated by it taking business media asleep at the wheel, or worse still, that wanted these companies to win and help them by lying. I have done a comprehensive evaluation of the last three years of press around chat, GPT and GPT itself, and you will look in the things in twenty twenty three and there's shit that's just fucking made up. There's a

whole go and look up task Grabbit GPT four. There are so many I'm a link one in the fucking notes. There are people that claim that GPT four ordered a task grab it to complete a capture. Now, on top of this not being a thing that task grabbit generally it does. This is from the system card of GPT four and it claims that it hired a taskrab, except when you look at just said it messaged them. It's very clearly made up. But everyone reported it as agents

existing in twenty twenty three. Aha. Every time I read this stuff, I feel little goddamn insane. But anyway, because these models do not have obvious replicable ways outside of benchmarks of testing what they can do, each user is effectively in a constant vibe check with the models, and the sycophantic qualities of GPT four oh were clearly enough

to endear them to the platform. People using GPT four Oh, they couldn't tell you why it's different to GPT five other than it feels less human or doesn't do the same things, even if those things are kind of hard to define. This is what happens when you build a following for a product based on specious hype and vague promises and lines of inference, of course, and then allow

users to make up the reasons that they care. You begin engaging with the gamer mindset of vibes based fandom that's completely unbreakable unless you make one subtle change that you could never see coming that breaks the illusion, leading to gamer like distrust and anger. You see, I theorized that the vast majority of chat GPT users do not

know where they used it in the first place. Three years of immediate pressure to use AI that AI was the future, their boss is saying AI is important and that you would be left behind if you didn't use AI. Mean that people come to chat GPT to work out why they're using it in the first place, which has led to all sorts of bizarre emotional attachments, kind of like one's attachments to a live service game and shout out to Catharsis twenty three on Blue Sky, who made

this observation. As a result, users were incredibly sensitive to changes like removing or changing a model because their association with chat GPT was based on However, GPT four zero works and sounds by ripping it out and replacing it with GPT five, users immediately felt jilted and swindled by open AI, and much like a dying live service game, any changes that have been made as a result were

met with paranoia and confusion. Clamiel Altman's attempts to paper over the problem by boosting rate limits on GPT five's thinking and restoring access to GPT four to ZH and other models for pay maid users were not enough because, in a very real sense, many of those users could not tell you what they liked four to oh to begin with four oh wasn't good so much as it

was an investment of time. By showing that open ai is willing to cut things arbitrarily, users can no longer trust that this investment of time is worthy, especially as many complain that it launch GPD five deleted a bunch of conversations. Now open ai sits in an odd spot where they're supposedly huge Manhattan Project level launch has been

met with either apathy or agony. While they've plicated users in the short term, it's very clear that the vast majority of users this, like GPT five and power users, don't seem particularly impressed with it either. This was meant to be the big launch that changed things for open ai forever, but it's turned into something of a mass betrayal or just kind of a mass letdown. And because it's based on vibes rather than its actual ability to do something, there's very little one could do to fix

this problem. It's unclear how all of this affects the company long term, but things do not seem good. Sam Wortman has already said that open ai is having to reallocate capacity for the couple of months prioritizing paying Chat GPT users of an API demand up from the current allocated capacity and commitments that they've made to their customers code for those who do not want to pay priority processing, which is now available for any developer. It's also unclear

what happens next. GPT five is not the future. Open ai is running out of capacity, and their product, despite the fanfare, has no capabilities or reasons to adopt it that are really new or interesting. Years of allowing the media to spin out ridiculous narratives about what AI can or could do using vague pablem that kind of suggests they're more powerful than they are has created a pr campaign for a product that does not exist, and the seven hundred million weekly active users of chat GPT have

clearly arrived there without much guidance. The attachment born of compulsion and societal pressure rather than any real use cases. When you allow people to define an indeterminately powerful tool by any standard they like, with no interest in correcting them, with no interest in guiding them, with no interest in actually showing them what it was that they were paying for others. Then it can generate staff, You'll create an attachment to it that devis any real ability you have

to control things. Open ai was never forced to productize and at scale. It's a very real possibility that people have pressure by the media and society itself force themselves to find meaning in LMS somehow, even if it feels kind of stupid. And what I mean by this is you get to the product and there really isn't that much guidance. Go on an open AI's website and have a look at the chat GPT page and look at

what it tells you to do. It's quite vague. You can look at it can analyze data, right, it can generate stuff. It helps you do ideas. Is that good? Am I smart for using this? Everything else you look at in the software world will tell you what it is you're using it for any consumer driven software at the very least. Yeah, chat GPT's never had to because the media for three years or two years, I guess with chat GPT has kind of just sat there doing the work for them, telling them, oh, yeah, you can

use it as a powerful person a system assistant. How to assist me with what? Nasty? Kevin Russe creepy Kevin in the New York Times a few weeks ago, when he did Everyone's Using AI piece with Casey Newton, he said that it's a powerful assistant. It's like, for what you can't do that? This shit can't control my calendar. I don't want to touch in my emails. I don't

think most people do either. So you've just got people to use it as a shit, our search engine, an online companion, and a brainstorming thing, which is a natural way to get people kind of addicted, but addicted to a product you don't truly control and a product you don't truly understand, one that can be swiped by just about anyone. I actually think we're on the the kind of downward spiral for this shit. I am winking like a pig and squawking like a bird watching this happen.

Even core Weavers crashing as I record this. They're down seventeen point nine one percent. Still up too much, though, And I really do think something has shifted thanks to GPT five, I'm a excited and like Doctor Stones once said, get excited because I think the next few months, like next year, is going to be chuckle heavy's. We're going to be whimsy pilled as we go through the remainder of the AI Boom, and I'll be here to guide you through it. Thanks for listening.