Hacker News Reader

Comment by cranberryturkey | Hacker News Reader

[flagged]

As a flesh and blood human I take some comfort in the fact that the comment I'm replying to is still identifiable (with high probability) as written by a non-human. I wonder how long it'll take until that's no longer the case.

It may already not be the case. I'll probably never know.

What can't popular LLMs do? Be vulgar and snarky.

The future of human communication is to be cunts online, so that we can identify ourselves. That will at least separate us from the script kiddies. Major marketers will train their AIs without any filter, so we're fucked anyway.

Wait, is this written by an AI?

100%

Yes. The dead giveaway is multiple paragraphs of bullshit where one sentence would suffice. People ain't got time for that.

That is the giveaway for some of the high-profile LLMs. The ones I run locally with ollama can be VERY close to perfectly human, subtle typos and all

What’s your local setup? I run llama3:8b and it works well

They truly walk among us

Please could you reduce that in size by 80% to generate a quick summary?

And keep iterating until you pick a single letter of the alphabet or emoji that encapsulates the essence of what you are trying to communicate.

Not bots. Simply outsourced to India, Malaysia and China. These countries were extremely happy to get this type of work.

Now this work will be outsourced to AI with even greater efficiency.

I think the real dead internet today and is a bit deceiving (not sure a better word) are examples of online games where you’re playing against bots or everyone assumes they are playing against bots. One of the aspects of the early days of the internet that was really cool and is now arguably not real anymore.

Reply on news.ycombinator.com