Yeah… and if they did figure something out, it would likely go into a AI detection tool rather than being shared openly… Found the source…
Who Broke The News?
Scott Aranson, a computer scientist working for OpenAI for AI safety and alignment, reported at one of his talks at the University of Texas that the team is working on cryptographic watermarking of the AI-generated content from Chat GPT to develop a distinction between human-generated and AI-generated content.
This is how Aranson puts it:
“For GPT, every input and output is a string of tokens, which could be words but also punctuation marks, parts of words, or more — there are about 100,000 tokens in total.
At its core, GPT is constantly generating a probability distribution over the next token to generate, conditional on the string of previous tokens.
After the neural net generates the distribution, the OpenAI server then actually samples a token according to that distribution — or some modified version of the distribution, depending on a parameter called ‘temperature.’
As long as the temperature is nonzero, though, there will usually be some randomness in the choice of the next token: you could run over and over with the same prompt, and get a different completion (i.e., string of output tokens) each time.
So then to watermark, instead of selecting the next token randomly, the idea will be to select it pseudorandomly, using a cryptographic pseudorandom function, whose key is known only to OpenAI.”
OpenAI engineer Hendrik Kirchner has also developed a working prototype for it.
They came to same conclusion…
“Now, this can all be defeated with enough effort.
For example, if you used another AI to paraphrase GPT’s output — well okay, we’re not going to be able to detect that.”
So, unless OpenAI has not changed the watermarking plan, as Aranson explained it, all those using the AI-generated content are in luck, but they would need a bit of effort.