Hacker News Reader

Comment by bee_rider | Hacker News Reader

It will be quite weird to have this brief blip in the anthropological record of like 1980-2010 where cameras were cheap enough that lots of real images were captured, and then 2010-2019 where everyone (in rich western countries at least) had cameras all the time and we got tons of real images of everyday life (filters are popular of course but using them is an intentional choice).

Then everyone goes inside for like 2 years (pandemic), there isn’t much to take pictures of, and then by the time we start going back outside everything is actually paintings by AI artists.

View on HN

When the AI generated content outnumbers human content 10,000 to 1, and the quality is almost ridiculously varied and sometimes incredible, and no functional system can sustainably authenticate humans without being overrun by bots…

Then the record from that point forward and forever hence will be one of AI, not one of humanity.

Arguably this is already true of text. We have already entered the age where it is impossible to determine with certainty whether a blog article was written by a human, and so every text data set that includes 2023 internet data will include ChatGPT outputs as ground truth.

Future historians will likely place a great deal of emphasis on 2019 being the year in which we reached a forever-maximum of non-pandemic-tinged provably-human-created data. It's no reason to panic, and the mixed data sets will absolutely be representative of humanity's accomplishments as a whole... but there will be notable differences.

> We have already entered the age where it is impossible to determine with certainty whether a blog article was written by a human, and so every text data set that includes 2023 internet data will include ChatGPT outputs as ground truth.

Is that any worse than unreliable humans? Whether they're intentionally lying or just mistaken, internet posts should always be taken with a grain of salt, for decades now.

The unreliability of humans is still valuable data because it shows what humans were thinking at the time.

I guess, but that's a really noisy signal once you account for deliberately misleading posts.

Even deliberately misleading posts give you insight into what people are thinking at the time, though.

I think they mean ground truth in terms of what human created content looks like, not ground truth in terms of what is actually going on in the world, which of course we don’t collect completely accurately.

I wonder what will be the first year when most of the meaningful comments on hacker news will be AI generated.

I dunno, seems hard to say. Upvote systems favor funny quips and matching the popular sentiment so I guess we won’t get a good measurement of meaningfulness.

For historians its a catastrophe. After 2023, the good old roman source poison assumation is back in full effect. He who is in power can actually rewrite history. Assume everyone uses written documentation just to proof loyalty to the current computational power holder.

Truth will become something similar to the roman emperors history, a million comments, blog articles condemning past rulers and CEOs as wrong all along, while praising the current instance, with similar spam in the forgotten archives.

A species ruled by madmen, one worser then the other. Praise be on the Bezos, who undid the mad world that reigned before.

> Is that any worse than unreliable humans?

Yes, I think so. Maybe not (but maybe) in terms of factual information, but in terms of opinion, analysis, arts, etc. You know, human things. The value of those things is, in my opinion, tightly couples with them being produced by humans.

I fear the social effects. The days of making long distance friends or relationships over the internet are over. Anonymous communities that don't implement strong KYC are over.

What is KYC?

Know Your Customer

It may be a great thing. We have in-person events vs. completely fabricated online wonderland. It might - I hope - mean that people flock back to IRL activities.

Which isn't great for those of us with very little ability to attend activities outside of the home.

OTOH, I have several very deep, meaningful and valuable friendships, spanning decades, with people I've never been physically near. Losing the ability to develop those sorts of relationships is a very real loss.

And I have never stopped seeing my IRL friends in person. Those, I almost never talk with electronically.

It's not that hard to filter content by source. Reputable news sources and authors will not publish AI-generated content (without grave risks to their reputation). And most of the abundant background noise have never provided much value before either.

> Reputable news sources and authors will not publish AI-generated content...

This news sources are not just one blob or a black box. It's different people that are working for money. And we know reputable sources that have published fictionary stories as real over years(see Spiegel in Germany). Now it's really easy to unnotable use AI, so where is the reason why this will not happen?

cnet published a bunch of unlabeled ai-generated articles for months https://www.theverge.com/2023/1/20/23564311/cnet-pausing-ai-...

the nyt just published a long interview with sydney in which she attempted to convince the author his wife didn't love him, which is ai-generated content; while that was clearly labeled, it will still appear in text data sets as ground truth

abundant background noise provides a lot of value for text data sets because it tells you how people talk

In all fairness, cnet hasn't been a reputable news source for a very long time.

I’ve yet to have ChatGPT produce useable content for my technical writing, or even lighter portions of proposal writing, rec letters, emails etc. I found it has done well at cleaning up text, fun language games, and quickly sourcing readily available information. I don’t think 2023 is the year books are replaced (maybe low novelty blogs). I feel a disconnect with how GPT is being exalted, and my underwhelming experience for useable content.

I suppose the veracity of history could be an excellent block chain application.

Or…art and literature enter a new two-tiered era with artistic and literary works made by humans, with other humans as witnesses to some or all of the process, becoming extreme-premium-items alongside AI-generated works, which are likely to become the fare of mainstream society, if the current wave continues unabated.

> everything is actually paintings by AI artists

It just occurred to me that we're actually going through the opposite transition outlined in Pratchett's discworld. Oh the irony.

> Down in the cellar Otto Chriek picked up the dark light iconograph and looked at it again. Then he scratched it with a long pale finger, as if trying to remove something. 'Strange . . ,' he said. The imp hadn't imagined it, he knew. Imps had no imagination whatsoever. They didn't know how to lie.

From Terry Pratchett's "The Truth"

I have the same thought about all artistic content. We've had a period of hundreds of years where we were advanced enough to reliably record, preserve and duplicate creative works but not so much that they were influenced by camera tech, photoshop, generative AI, LLMs and all other digital craziness. This reliance will only continue to increase as the tech gets better. After a point will it be too difficult to ever produce anything unique? Is the golden age of human creativity behind us?

Reply on news.ycombinator.com