3 ms·

Given how many of those SEO spam sites are themselves generated by ChatGPT now, OpenAI can simply back-reference their own logs to find out which sites are prob

Given how many of those SEO spam sites are themselves generated by ChatGPT now, OpenAI can simply back-reference their own logs to find out which sites are probably SEO spam while everyone else is left guessing. That's vertical integration!

arromatic 2y ago
If they do that , That's a genius idea.
- code51 2y ago
  So it'll turn to yet another arms race - similar to captcha, cybersecurity and nuclear weapons. SEO will use AI to fill in fluff inside AI-generated content (which is already done). It won't directly match ChatGPT logs and OpenAI would just be pouring precious compute to a bottomless pit trying to partial-match.
  - jerjerjer 2y ago
    Serve Claude-generated version to OpenAI bots. Serve OpenAI-generated version to Claude bots. Problem solved. Serve users a random version and A/B test along the way.
    - methyl 2y ago
      Then you are still left with self hosted models which are pretty good at this task.
DSingularity 2y ago
I’m sure they will be more subtle than that otherwise it will get circumvented. I’m sure they will/are tackling this at the model level. Train them to both generate good completions while also embedding text with good performance at separating generated and human text.
- sebzim4500 2y ago
  Would someone even want to circumvent it though? Most sites won't care very much about encouraging scrapers to include them in LLM training data, it's not like you get paid.
  - DSingularity 2y ago
    If your website is created to promote a product of course you are incentivized to be included.
- LegionMammal978 2y ago
  Personally, I'm a pessimist on this front. People assert that a model-in-training can effortlessly sift out the real data from mountains of LLM spam. But then people also assert that AI detectors do not work and can never work, since LLM output is simply too good, and any watermarking can be broken up by a light paraphrasing step. It doesn't make much sense to have it both ways. I can only await companies' attempts to publish enough junk to create an 'alternative truth' for new LLMs to believe in. The worst part is, it might even work.
itissid 2y ago
Or offer two search results when they suspect one is spam and see which one a user likes and train off of that, just the way they do now with ChatGPT.

arromatic 2y ago

If they do that , That's a genius idea.

code51 2y ago
So it'll turn to yet another arms race - similar to captcha, cybersecurity and nuclear weapons. SEO will use AI to fill in fluff inside AI-generated content (which is already done). It won't directly match ChatGPT logs and OpenAI would just be pouring precious compute to a bottomless pit trying to partial-match.
- jerjerjer 2y ago
  Serve Claude-generated version to OpenAI bots. Serve OpenAI-generated version to Claude bots. Problem solved. Serve users a random version and A/B test along the way.
  - methyl 2y ago
    Then you are still left with self hosted models which are pretty good at this task.

DSingularity 2y ago

I’m sure they will be more subtle than that otherwise it will get circumvented. I’m sure they will/are tackling this at the model level. Train them to both generate good completions while also embedding text with good performance at separating generated and human text.

sebzim4500 2y ago
Would someone even want to circumvent it though? Most sites won't care very much about encouraging scrapers to include them in LLM training data, it's not like you get paid.
- DSingularity 2y ago
  If your website is created to promote a product of course you are incentivized to be included.
LegionMammal978 2y ago
Personally, I'm a pessimist on this front. People assert that a model-in-training can effortlessly sift out the real data from mountains of LLM spam. But then people also assert that AI detectors do not work and can never work, since LLM output is simply too good, and any watermarking can be broken up by a light paraphrasing step. It doesn't make much sense to have it both ways. I can only await companies' attempts to publish enough junk to create an 'alternative truth' for new LLMs to believe in. The worst part is, it might even work.

itissid 2y ago

Or offer two search results when they suspect one is spam and see which one a user likes and train off of that, just the way they do now with ChatGPT.