Winter Rant

Reading the responses that we are getting from GPT and its likes (Bard for instance), is like staring at ourselves in the mirror.
I am specifically talking about the so-called imaginary nonsense these systems are capable of spewing out, which i am guessing is in large part due to the training data used to build these models. Experts in the field call them "hallucinations" … almost as if these models have a mental state, or a mind to speak of.

But i briefly turn my attention to where these hallucinations came from …

I think back to all those useless bullet-listed articles that masqueraded as journalism (thanks Huffington Post!).
I think back to all the untested, buggy open-source code repositories that we collectively saddled on Github (of which i am probably responsible for about 100 repositories).
I think back to all the hate, and vitriol on image boards.
I think back to all the clickbait.

Those parts of the internet likely made their way into the training sets for these large language models — at least I have no reason to think that they did not.
And so, when I read some really crazy responses that ChatGPT and its ilk keep spewing back at us, I cannot help shake the feeling that we (humans) are looking at a reflection of our collective existence on the internet of about (nearly) three decades.

I guess you sow what you reap.