Hacker News Reader

Comment by anaganisk | Hacker News Reader

When the beta came out I was very surprised to see overnight startups bad on it. Some even hiring devs. I still wonder how they thought it was a good idea to launch because one sparse example worked, without any control or knowledge over internals and pricing.

View on HN

There is a case to be made for being fast "online" with a new breakthrough technology. So they just made their bet as any startup bets on a central thesis.

And its not like GPT-3 is useless. It is perfect for immitating texts written by other entities.

Cynically, it'll probably end up being used to generate content for websites and online marketing.

It already is, and I regret to admit I've used it.

Check out https://copysmith.ai or https://shortlyread.com.

Are the results any good? Can it really generate quality copy? How is it for research on any given topic?

I wonder if English will be the first to become a low trusted language as most AI research is done on the English language. “Is it in english? Probably written by a bot.” might be a common thought in ten years. Ofc eventually the big languages will catch up in marketing spam.

If this future happens, people might "fall back" to their native languages for internet writing. I'm already thinking about this: if I wrote something which could potentially get me in legal trouble, like a reverse engineering post, I would probably write it in my native language. Kind of similar to torrent trackers in Russian or obscure phone mod forums in Portuguese.

Even worse, armies of GPT-3s probably wage massive flamewars against each other all over the internet in an attempt to sway public opinion.

It was always inevitable though.

Yeah. And while GPT-3 output is often nonsensical, it's nonsensical in the same way a good fraction of the general public writes on the Internet. In this way, it's close to passing the Turing test - not because its good, but because real humans on generic platforms[0] are really that bad.

[0] - By that I mean platforms that aren't strongly moderated for quality of discussions, or that don't focus on niches for which the set of people interested in them already has a higher than average discourse quality.

Locke versus Demosthenes ...

quite many startup exist to capture vc capital, not market share

Even more insane is that anyone can replicate their product if they figure out a similar primer.

GPT-3 startups have practically no moat in my opinion.

they could automate putinbots/web brigades, sure there must be a big market for automating politically motivated spammers. Or bots that write product reviews that seem to make sense, you could find applications wherever someone would like to create the illusion of public discourse (everything is weaponizable). Each of these 'applications' would have to tweak the modell, somehow.

Reply on news.ycombinator.com