I believe “their own language” is popular miscommunication of the Gibberlink demonstration from February last year where LLMs identify each other and switch to ggwave communication. I don’t know of any evidence for models being “purged” as they claim.
This is an ad for facebook’s chatbox. I am a computer scientist. The chatboxes pointed at each other are poisoning their own datasets by optimizing for what the average human conversation appears like, but they are no longer talking to humans. Look at the example used in the article.
Bob: “I can can I I everything else”
Alice: “Balls have zero to me to me to me to me to me to me to me to me to me to”
Without a human involved, the chatbox is useless, as it can’t teach itself what is being expected of it. It’s essentially “”““learning””“” that nonsense is an acceptable output.
I don’t want to doubt Mx. @trezzscience, but this is a really funny article. Re-read the nonsense above and tell me youre scared of it.
Well they did purge a few AIs because they started inventing their own language and evolved too fast so yeah i see your point.
I don’t believe your source is a reliable science communicator, but regardless I’d like to know more about this.
https://medium.com/@trezzescience/the-ai-that-invented-its-own-language-and-why-scientists-shut-it-down-aa92c94466bc
I believe “their own language” is popular miscommunication of the Gibberlink demonstration from February last year where LLMs identify each other and switch to ggwave communication. I don’t know of any evidence for models being “purged” as they claim.
https://medium.com/@trezzescience/the-ai-that-invented-its-own-language-and-why-scientists-shut-it-down-aa92c94466bc
I think you need a break from the internet, friend.
https://medium.com/@trezzescience/the-ai-that-invented-its-own-language-and-why-scientists-shut-it-down-aa92c94466bc
This is an ad for facebook’s chatbox. I am a computer scientist. The chatboxes pointed at each other are poisoning their own datasets by optimizing for what the average human conversation appears like, but they are no longer talking to humans. Look at the example used in the article.
Without a human involved, the chatbox is useless, as it can’t teach itself what is being expected of it. It’s essentially “”““learning””“” that nonsense is an acceptable output.
I don’t want to doubt Mx. @trezzscience, but this is a really funny article. Re-read the nonsense above and tell me youre scared of it.