INFERENCE COSTS NOW THE LIKELY MURDER WEAPON FOR THE DEAD INTERNET
It's widely agreed that GPT5 was a disappointment with little quality improvement over GPT4. This suggests LLMs are already approaching an ability ceiling. However, that doesn't mean the gig is up....
The current state of the art is already at a quality level that can produce writing indistinguishable from human. There's still room for improvement due to some quirks like the em-dash issue, but these seem solvable without major advances.
So, if chatbots can now imitate humans effectively, why isn't the internet dead yet? Why haven't we yet seen total domination of bot replies and accounts on social media?
The answer is inference costs. Mass producing replies still isn't free and instantaneous. That will change.
Folks, I'm going to have to boomer you because it's the truth. Lots of the problems forming relationships are caused by digital socialization. If we dropped you all into a school dance on a desert island with no phones and only coconuts and fish to eat, you'd pair up in a month.
Twitter is especially bad due to Darth Algorithm ("feel the hate flowing through you [and post]") promoting divisive rhetoric. But, IG creates problems of its own--it encourages lifestyle chasing to a degree never before seen. Influencer is a four-letter word on any platform.
The cues that attract you to the other sex in a romantic way are mostly NOT PRESENT in the digital world. And, it has all sorts of other disadvantages as well, which I've covered before. For instance, popularity follows a power law. Physical space sets a limit so no power law.
Why ugly things like millennial gray attached garages and beige lingerie are in style and unstoppable by rational and aesthetically aware criticism even though they're completely retarded:
While the last stage, "people finally get bored," does happen sometimes, it's often OVERLY OPTIMISTIC and in reality bad trends can go on until we're all dead.