5 ms·

(OP here) OpenAI recently demonstrated a frighteningly high quality text generator[1] which leaves me concerned about the future.. humanity of the web - no mor

(OP here) OpenAI recently demonstrated a frighteningly high quality text generator[1] which leaves me concerned about the future.. humanity of the web - no more of those sub-par English articles if a bot writes for you. Makes me think we'll inevitably see large tracts of the web near-fully automated - information gathering (research) and text generation (writing) for articles based on a simple prompt for that spark of creativity. Next ten years sure will be interesting on the web. [1] https://techcrunch.com/2019/02/17/openai-text-generator-dangerous/ https://techcrunch.com/2019/02/17/openai-text-generator-dang...

bhartzer 7y ago
Computer-generated content has been around for a long time... and I even tested creating websites with auto generated content in the 1990s. It's nothing new. In fact, the content has been so good for a long time that even scientific papers get accepted https://gizmodo.com/mathematics-journal-approves-paper-filled-with-computer-5953147 https://gizmodo.com/mathematics-journal-approves-paper-fille... and that was back in 2012.
- Swenrekcah 7y ago
  That was an example of a broken review process, not of good content.
Bombthecat 7y ago
Who says, that your comment isn't a bot already? Internet is really turning into a low trust society... I guess network sites with only your friends will become more important then. With probably even bigger niche conspiracy and antivaxx bubbles then now..
- dorgo 7y ago
  >Who says, that your comment isn't a bot already? Why do people care whether a human or a bot wrote a comment / draw a picture / developed software.. ? The only relevant metric is the quality of the comment / picture / software, isn't it? https://xkcd.com/810/ https://xkcd.com/810/
  - cameronbrown 7y ago
    Because humans are social creatures. Talking to a bot, even an incredibly accurate one, isn't the same. Tech like deepfakes or generated text is going to require us all to cryptographically sign content in the future. I foresee a time where white house press releases must be signed by the president just to confirm its real content.
    - KingFelix 7y ago
      Beep Boop Beep...your comment is great
    - whalabi 7y ago
      I've wondered if I should make a Reddit where everyone proves their ID, maybe even with a video so it's a lot harder to fake, so everyone can only have one account, and you'd know they aren't people working for the Russian government or whatever.
    - bennyelv 7y ago
      That doesn't quite solve the problem unfortunately. If this is the only solution then the subject of some content can repudiate it by not signing it, even if it did actually happen. For example if the president was filmed saying something in public he later wanted to deny, he could just claim it was faked. The person who filmed it could sign it, but who are they and what authority do they have to claim that they didn't fake it?
  - dredmorbius 7y ago
    That is turning out to be one of the most misguided xkcd comics ever. The problem is that bot traffic is virtually definitionally not neutral; it is trying to attempt some goal, overt or covert. That might be advertising, link farming, submarine placement, karma whoring (to some later use), vulnerability probing (if only for anti-bot defences), information gathering, or various forms of propaganda: disinfo/misinfo, instigation, or just overwhelming channels with noise (the Chinese model). Even putatively benign and declared bot rapidly become at best annoyances as experiences on Reddit and Mastodon are showing. Good bot practices require very judicious and closely administered use. Wikipedia and IRC service and janitorial bots are examples of positive use. Each is controlled and deployed by specific administrators. I've seen and reported appearance of covert bot accounts across multiple networks. They never prove benign. All of these reduce the human-to-human trust of a network -- the presumption that a given interaction is with an actual human, absent grossly obvious ulterior motives. As a few others have noted, excessively mediated experiences seem to become lower rather than higher trust, a characteristic I've suspected for some time is a generalised case, with some possibly exceedingly deep cultural and sociological implications. This rapidly veers off into my own hypothesis and theories, but; looking back at histories of communications technology and practice, particularly over the past 200 years or so, the practice of imprved coms is parallelled by routinised and standardised formats, and far less obsequious and religious language. (See especially research by James Beniger and JoAnne Yates on commercial and business communication). As comms latencies fell (from weeks or months to days or minutes, or less), the need to specifically prove and swear allegiance, loyalty, thankfulness, and debt in every exchange falls. I also suspect (though I've found no discussion) that a major role of empirical religious traditions, which every major preindustrial empre seems to have had, is building a shared social trust basis. This would be a social mechanism of religion. Other interesting observations are that it was the greek god Hermes represented trade, merchants, commerce, roads ... and thieves and trickery. It was as the first high-spped comms networks (telegraph) and motorised highway system emerged (Mississippi steam-powered river boats) that the idea of the confidence man emerged in Herman Melville's novel of the same name. The elements of distant communications, and speedy entry and exit, create the space required for fraud. A different tack is taken by Kyle Harper's The Fate of Rome, looking at the co-emergence of disease, epidemics, and plague with both urban density and high-volume (for the time) trade, commerce, and travel. Though he doesn't touch on fraud. The whole question of how and why high- and low-trust cultures develop is an interesting one.
CM30 7y ago
Honestly, that'd probably be better than some of the blogs you see in Google's results at the moment. At least the sample text provided there was readable.
- nradov 7y ago
  They ought to introduce some random spelling and grammar errors to make it more realistic.

bhartzer 7y ago

Computer-generated content has been around for a long time... and I even tested creating websites with auto generated content in the 1990s. It's nothing new. In fact, the content has been so good for a long time that even scientific papers get accepted https://gizmodo.com/mathematics-journal-approves-paper-filled-with-computer-5953147 https://gizmodo.com/mathematics-journal-approves-paper-fille... and that was back in 2012.

Swenrekcah 7y ago
That was an example of a broken review process, not of good content.

Bombthecat 7y ago

Who says, that your comment isn't a bot already? Internet is really turning into a low trust society... I guess network sites with only your friends will become more important then. With probably even bigger niche conspiracy and antivaxx bubbles then now..

dorgo 7y ago
>Who says, that your comment isn't a bot already? Why do people care whether a human or a bot wrote a comment / draw a picture / developed software.. ? The only relevant metric is the quality of the comment / picture / software, isn't it? https://xkcd.com/810/ https://xkcd.com/810/
- cameronbrown 7y ago
  Because humans are social creatures. Talking to a bot, even an incredibly accurate one, isn't the same. Tech like deepfakes or generated text is going to require us all to cryptographically sign content in the future. I foresee a time where white house press releases must be signed by the president just to confirm its real content.
  - KingFelix 7y ago
    Beep Boop Beep...your comment is great
  - whalabi 7y ago
    I've wondered if I should make a Reddit where everyone proves their ID, maybe even with a video so it's a lot harder to fake, so everyone can only have one account, and you'd know they aren't people working for the Russian government or whatever.
  - bennyelv 7y ago
    That doesn't quite solve the problem unfortunately. If this is the only solution then the subject of some content can repudiate it by not signing it, even if it did actually happen. For example if the president was filmed saying something in public he later wanted to deny, he could just claim it was faked. The person who filmed it could sign it, but who are they and what authority do they have to claim that they didn't fake it?
- dredmorbius 7y ago
  That is turning out to be one of the most misguided xkcd comics ever. The problem is that bot traffic is virtually definitionally not neutral; it is trying to attempt some goal, overt or covert. That might be advertising, link farming, submarine placement, karma whoring (to some later use), vulnerability probing (if only for anti-bot defences), information gathering, or various forms of propaganda: disinfo/misinfo, instigation, or just overwhelming channels with noise (the Chinese model). Even putatively benign and declared bot rapidly become at best annoyances as experiences on Reddit and Mastodon are showing. Good bot practices require very judicious and closely administered use. Wikipedia and IRC service and janitorial bots are examples of positive use. Each is controlled and deployed by specific administrators. I've seen and reported appearance of covert bot accounts across multiple networks. They never prove benign. All of these reduce the human-to-human trust of a network -- the presumption that a given interaction is with an actual human, absent grossly obvious ulterior motives. As a few others have noted, excessively mediated experiences seem to become lower rather than higher trust, a characteristic I've suspected for some time is a generalised case, with some possibly exceedingly deep cultural and sociological implications. This rapidly veers off into my own hypothesis and theories, but; looking back at histories of communications technology and practice, particularly over the past 200 years or so, the practice of imprved coms is parallelled by routinised and standardised formats, and far less obsequious and religious language. (See especially research by James Beniger and JoAnne Yates on commercial and business communication). As comms latencies fell (from weeks or months to days or minutes, or less), the need to specifically prove and swear allegiance, loyalty, thankfulness, and debt in every exchange falls. I also suspect (though I've found no discussion) that a major role of empirical religious traditions, which every major preindustrial empre seems to have had, is building a shared social trust basis. This would be a social mechanism of religion. Other interesting observations are that it was the greek god Hermes represented trade, merchants, commerce, roads ... and thieves and trickery. It was as the first high-spped comms networks (telegraph) and motorised highway system emerged (Mississippi steam-powered river boats) that the idea of the confidence man emerged in Herman Melville's novel of the same name. The elements of distant communications, and speedy entry and exit, create the space required for fraud. A different tack is taken by Kyle Harper's The Fate of Rome, looking at the co-emergence of disease, epidemics, and plague with both urban density and high-volume (for the time) trade, commerce, and travel. Though he doesn't touch on fraud. The whole question of how and why high- and low-trust cultures develop is an interesting one.

CM30 7y ago

Honestly, that'd probably be better than some of the blogs you see in Google's results at the moment. At least the sample text provided there was readable.

nradov 7y ago
They ought to introduce some random spelling and grammar errors to make it more realistic.