The idea raised in this article is interesting. However, if the technology on the side of the spam producers improves, platforms will follow and improve their technology for recognizing spam.
So, yes, there might be some temporary lemons failures on some platforms, but I would not expect a big logging off to happen on the long-run. After all I think this is a cat-and-mouse game of superior technology in producing / identifying spam.
> platforms will follow and improve their technology for recognizing spam
Written language has patterns, and your response and mine also follow these patterns. Language models like GPT-3 and ChatGPT can learn these patterns because there are finite ways in which humans write text. ChatGPT responses are easy to recognize because it was fine-tuned with specific data to meet chatbot requirements and provide Q&A style responses. In the near future, these language models will be fine-tuned for specific purposes, such as spam, and will mimic human text patterns in a way that will not be distinguishable from real humans.
While all of the above are true, there is a variable which was a constant for thousand of years. That is that language is generally static, it doesn't change much over ten, twenty years, or fifty years. It sometimes undergoes significant changes over the centuries.
Well that constant will be transformed to a variable. While GPT can recognize and produce outputs of a language, this tool, or similar tools can be used to produce another language, totally from scratch or a variant of an existing language, instantly.
In the language context, i would argue it is a lot more difficult to produce effective spam than in the image context or video. People will change their language, the model will have to be trained again in the new language, and that is until people up their game and change language again. This is happening already, young people use different language in the internet than older people. I would be surprised if we don't witness in the near future, the internet of a million languages.
How many people on Hacker News do you know, or is that just a way to say you're looking for work right now?
People here have experimented with using GPT-3 to generate comments and content shared here, well before chatGPT was released; while it can sometimes be a bit suspicious, it can also get up-voted: https://www.theverge.com/2020/8/16/21371049/gpt3-hacker-news...
You didn't answer my question. I barely spend more than 15-20 mins a day on HN, so how does it matter if I am reading something useful or not. This is just an imaginary problem like the problem of having 100 million vids of how to boil and egg on youtube. Once you get the info you need from one vid who sits thro the rest. People just tune it out. Like traffic or bar conversations.
> How much time in day, do you think you spend reading what random people you dont know are saying?
Rather than "so what?", which I will come back to later: For me, probably several hours per day. It's not just here, or twitter/fb, though that's probably an hour by itself, it's also:
• Blog authors (parasocial relationships!)
• Newspaper writers
• Wikipedia editors
• Review writers on Amazon, Google maps
• Question and Answer writers on Stack Overflow
• The first few comments under some Youtube videos
• Comments sections on news stories
(Technically also book authors given the way you phrased that, but that's a different issue and current generation text generators make no difference there anyway).
Now assuming the question was "so what?", because that dovetails into the rest of your spiel:
The problem is not usefulness to you, it's "can this be used to automate convincing you?".
Make a half-decent sales pitch personalised at whoever for whatever: "Boiling eggs? Yeah that's easy, but I find organic is the best and well worth the cost", or "Oh, you're visiting Berlin? Yeah, I had a really great time at $insert_restaurant_here" or "Don't fall for the Lib Dems, they keep saying they can beat the Tories but look what happened when they got into power. That's why I vote $insert_uk_political_party_here". All personalised to just you, in the form of a friendly looking and natural response that probably seems more natural than any of my examples, because chatGPT is only reliably worse than domain experts and I'm not a domain expert in sales.
Also: You replied to me, and I'm a stranger, so we both know you engage with strangers on the internet, QED there's a place where you might get a bot trying to sell you a product or idea while pretending to be human.
Don't make the mistake of thinking the status quo is a good reference point here: Current advertising categorisation on social media sucks (easy way to find out: make GDPR request to FB or Twitter for all the data they have about you. They think I'm interested in languages I can't speak and politics of countries I don't live in).
