I sometimes have to support Windows machines.

When I Google things for Windows that require 3 bullet-point sentences to answer, inevitably I get a 12-minute video of an Indian guy with an accent that prevents me from watching at 2x audio comprehensibly, taking 6 minutes to tell me how common the problem is and 6 minutes teaching me how to download and install spyware that does what I want and much more.

I love when you google for something and in order to stretch the content out 80% of it is completely unuseful, barely relevant, content.

"How do you change the sandpaper in a sander" and most of the article is explaining why sander X is the best sander on the market.

I wonder if the OpenAI models have video transcripts in their training.

Or if GPT-4 or -5 (whatever upcoming model) will understand video or visual information and it's relation to text.

Over the next few years, with grounded language understanding and other capabilities, no one will be able to pretend that these systems aren't intelligent. I mean, some people always will, but it's going to be a very small percentage. Right now I am guessing we are about 50-70% of people convinced that this stuff is cognitive rather than regurgitative.

But also I think the abilities of these models clarify the nature of intelligence and the relationship between intelligence, compression and computation.

Yeah, the hobbyists that used to make niche sites on subjects they loved have all moved to youtube for publishing their content.

They were the people who made the web truly great back in the day.