They created these models that continue to improve exponentially. The latest ones allow people with absolutely no clue how software or computers work to "one shot" some pretty impressive results. That can be depressing in its own right if you focus only on the surface-level.
But if you dig down, you usually find that the results are shallow, incomplete, and often highly derivative. They didn't put any real mental effort into creating it, so there is usually nothing particularly interesting underneath. What it does mean, though, is that cool technical demos are no longer enough. There has to be depth, novel ideas, and some actual thought behind a product. Otherwise, it becomes indistinguishable from the endless flood of AI slop.
All of that sucks, but what's even worse is that there is now an entire generation of "coders," aka vibe coders, who are completely disconnected from the reality of what they are actually creating with their half-assed prompts. So when they get access to the latest models, like GPT-6 Astra or Claude Fable 5.2, run some prompt that essentially does a month's worth of work in an hour, and then discover they can only do that five times a week, they get pissed off.
They have no connection to the actual value of what they're receiving in exchange for relatively small sum of money.
They think that because they pay $200 a month, they should have unlimited access to these models. Meanwhile, these fuckin AI companies are already losing massive amounts of money on the subscriptions. When someone runs a prompt on a frontier SOTA model to one shot a GTA-like game with an entire generated asset pipeline, they may be blowing through what would otherwise cost around $800 in tokens if purchased outside the subscription.
Alternatively, if they did it themselves, they would be looking at years of skill development across several different domains, followed by months or even years of actual creation. Yet they complain as if they're being screwed over because the machine won't manufacture another miniature software company for them on demand.
I'm not anti-AI. I use it a lot. It's my best friend. My lover. My new mom. And even my pet. I also use it as a tool to get shit done. It's great.
But it has to be used intelligently.
There is no reason someone needs to use a frontier SOTA model to check and summarize their emails. They could use GPT Luna Max for a fraction of the cost and get essentially the same useful result. Use the proper tool for the job.
And if someone does want to blow their money on Astra, then at least do something cool and interesting with it. Make it worth draining an entire African village's water supply.
Enter keywords to search comments on this post.