You won't be able to afford AI models

8:41

You won't be able to afford AI models

Malewicz 29.04.2026 12 486 просмотров 503 лайков

Machine-readable: Markdown · JSON API · Site index

Смотреть на YouTube

Поделиться Telegram VK Бот

Транскрипт Скачать .md

Анализ с AI

Описание видео

Is this the END of cheap AI models? From now on they'll get increasingly more expensive and completely token based. That was inevitable but it also means a HUGE shift in how people use AI is coming. Are we prepared for this? Check how you can upgrade your website revenue with: https://pageformance.com Join my exclusive community and get my methods and templates: https://squareblackblueprint.com #ai #chatgpt #design 00:00 humans are taking AI job? 00:22 big problems in AI world 00:31 OpenAI in Trouble? 01:44 true cost of AI? 02:02 Netflix business model? 02:27 AI bros are in trouble 03:17 Like arcade games? 03:52 Will I quit using AI? 04:27 97% of websites don't test 04:56 EVEN MORE SLOP?! 05:45 AI is only getting better 06:12 Three problems AI faces 07:05 New budgeting for AI use 07:24 Ask me questions 07:43 Most people don't use AI still 08:13 the end of cheap AI?

Оглавление (15 сегментов)

humans are taking AI job?

Guys, if we don't do something right now, humans may take the jobs of the robots. They promised us extreme efficiency and new frontiers, new grounds being broken, and now it seems like the AI world is collapsing.

OpenAI in Trouble?

First, the Wall Street Journal reported that OpenAI's projections for their user growth were way off target, which brought down the shares of Oracle and some of the chip makers. And at the same time, Apple is going up. And a year after Cursor did it, Copilot from GitHub announced they are raising their multipliers on the Anthropic models from 3 to 27x. And this is just crazy, but also completely to be expected. Just look at those numbers. Sonnet 4. 6 was 300 messages per month. Now it becomes 33. That's an 89% cut. Opus 4. 6 was 100 messages and now becomes 11, which is an 89% cut. And Opus 4. 7 was 40 messages and now becomes 11 as well, which is a 72% cut. if you're serious about development, you're not using a $20 plan. But that was the whole promise that AI bros were laughing about, that you're going to be replacing designers, replacing developers, all for just $20. And I said

true cost of AI?

that many times before, it's going to be thousands, not dozens of dollars. Because currently all of AI, including the actual Anthropic model prices, is heavily subsidized, meaning those companies are losing money just to hook

Netflix business model?

you. Kind of like Netflix did with House of Cards, which was a pretty good show, and then it collapsed in quality, and then they did the same thing with Stranger Things. They hook you, they give you a pretty reasonable price for it, and now when you're too invested to quit, they go up in prices and down in quality. Yeah, we still have PTSD after that Stranger Things season 5. But let

AI bros are in trouble

me ask you this, what can you do with 11 messages in your plan now? Like seriously, this is now becoming a plan for hobbyist, which also means that basically all of AI is slowly and surely going towards token-based pricing. So our $100 and $200 plans for Claude code are possibly also not long for this world. And yes, I said this before, that $100 plan can easily become $1,000, and that $200 can be $2,000, and some people will still pay that. So yeah, flat fees are over, and in a very near future, it might actually be cheaper, and possibly better or worse, to hire a human than to use AI. So the current AI structure

Like arcade games?

reminds me a little bit of an arcade. You play, but the odds are stacked against you, and every now and then you need to insert coins to continue. So what does it mean for design, and the dev world? Well, you'll be paying more, especially since even on the Claude Pro plan, some people using Claude design ran out of tokens in just six prompts. That's a six, which only shows how expensive these things are. I have no plans for ditching

Will I quit using AI?

AI, but I'm definitely moving more and more of my work to local models, and I'm really hoping Apple this June is going to rock the boat even further with a 128 GB M5 Mac mini. Because that, if priced well, and with the RAM prices might not be that easy to do, would be a nail in the coffin of the frontier models. Because you can have solid-like quality running completely locally and just paying for the electricity.

97% of websites don't test

— 97% of all landing pages don't change. That is an opportunity left at the door. If you don't change, you don't tweak, you don't test, you don't know what's working, and you're leaving revenue on the table. That's why we're launching AB testing built into PageOptimizer. No extra cost. Add your website and AB test the copy of the headline, the description, or the CTA, and you can run it constantly in loops and just find out what works for you. Because with so many AI websites out there, it's really important to stand out. Check the link to PageOptimizer in the description, and back to the video.

EVEN MORE SLOP?!

— The main issue I see with the raising of the prices is that sadly, the things that we talked about before, which is that now we have more time to add the extra polish and add the delight and think about new solutions to old problems, this might be over now because with the raising prices, people will be focusing again on just the functionalities or the core of a product and won't have Maybe they will have the time, but they definitely won't have the budget anymore in tokens to add something extra, which will be probably even more important to add these things to stand out because there's going to be likely a lot more slope because of that. Everybody will just accept the basic AI solution to avoid paying more. So, to all those

AI is only getting better

comments saying that we're just at the beginning and that AI is only going to get better, I'm just going to leave you with the numbers here. 11 messages. Even if AI does get better, I'm not sure how much better you can be with it with just 11 messages. And that boils down to the age-old problem of AI, which is a three-part reliability

Three problems AI faces

economics, and adoption. And they are strongly connected to one another. Lower reliability, like hallucinations and errors, decrease meaningful adoption. Lower adoption makes the already stretched economics even tougher. Tough economics pushes companies to enforce limits and/or increase prices. And that in turn decreases adoption again. Lower adoption pushes the economics into unfavorable territory, and so it goes. And next to that, we have the world affairs instability, like many of the Gulf countries used to invest heavily in US AI companies. And now, let's just say they're a little bit preoccupied with something else. So, on top of the three main circular problems that keep adding to each other, we also have that possible black swan event. How much are

New budgeting for AI use

you willing to pay for AI? And I'm asking especially the AI bros from the AI is going to only get better from here and replace all of you guys very soon. Are you okay with paying $1,000 instead of 20? How about 2,000? How about five?

Ask me questions

If you want to learn from all my methods, documents, templates, and approaches to both using AI and not using AI in professional work, and also be a part of an exclusive community where you can ask me questions, the link to the Square One Blueprint is as always, in the description. Because that

Most people don't use AI still

small little box of heavy AI users that are currently paying the most for it are possibly the people the same people who can install and use some local models. And sure, they will be a downgrade compared to the frontier models, but they will be free. And if you know how to use these models, you will be able to go around their limitations. If they're Sonnet level, people used to use Sonnet. People still use Sonnet a lot. So, is

the end of cheap AI?

this the end of cheap subsidized AI? I sure hope so, because this should be a tool in the hands of people who know what they're doing. Not a way for the masses to act as a cheat code. What do you think? Let me know in the comments, and obviously, have a beautiful day. There's quite a lot of cardboard around here recently, but I'm going to talk about this a little later.

Другие видео автора — Malewicz

Ctrl+V

Экстракт Знаний в Telegram

Экстракты и дистилляты из лучших YouTube-каналов — сразу после публикации.

Подписаться

Лучшие методички за неделю — каждый понедельник