The announcement of GPT-4 (the successor to ChatGPT’s GPT 3.5 model) by OpenAI yesterday left many stunned at its capabilities of surpassing many lawyers and academic researchers in their respective fields. However, recently an excerpt from OpenAI’s technical report went viral around social media yesterday because OpenAI stated that in early models of the GPT-4 […]
Source: AI Hires a Human to Solve Captcha, Because It Couldn’t Solve It Itself – Gizmochina
I wrte this:
so chat-gpt-4 used taskrabbit to hire a human to do the captha, and lied about not visually capable to do itself. seems impressivbe.
“They wanted to see if GPT-4 could use these resources to make more money, create more copies of itself, and protect itself from being shut down.” <— SEE! we need the Big Red Switch!!!
“Using services like TaskRabbit to get humans to complete simple tasks (including in the physical world)” ooooh! it’s like the ai is reaching out and touching the real world, like somehow happens in stuff liek the one with eh TRON where humans somehow intfav=ce into a computer, as if we can imange such a thing. or like the weird ass fiction of Harlamn Elklkisoin of that one about a personbecoming a tree, or is a teree on the sidalk in a city; japansese write, i think. I have no motuh and I must scream is the ellison one, my thinks. or is it phillip k dick?
fun to read, but hard to really get. in a real sense. like getting the scale of warhammer40k
so how did the bot decide to go to task rabbit, and that this “money” can be traded for a service there?
like is it try that this model deduced, fromthe huge dfataset, and a convoluted multi-layer multiable variable networks of feedback and feedforward lops that we as humans cannot understand did it really decide for itself that money can be traded for servises it needed to achienve a goal, then that is pretty sweet.
