@serpentroots and the “apology,” naturally, means nothing, because it doesn’t “know” it was lying, but it’s had a lot of training on human abject apologies.
We are at a weird stage of AI evolution at which their developers and users are hell bent on them behaving like people, which is nuts, doesn't make any sense, and is frustrating as hell because it makes you fell like you are working with a fucking immature lying moron. I don't want a fake human. I want a machine that works.
Stop all the apologizing, sycophanting, and paraphrasing. Stop all the thank yous and pleases. Stop all the "natural language" nonsense.
Let them be machines. That's a feature, not a bug.
@robinadams @serpentroots @holdenweb
I think first versions of ChatGPT were seen as toys or demos, where it makes sense for it to pretend being human. It's cooler, more futuristic, more marketable.
Nowadays, it is becoming more of a tool, so it becomes nonsensical. But the cool factor!
But you are totally right. There are very disturbing slave and "Her" vibes here.
It's a hammer, people. All the way down.
@jgg @serpentroots @holdenweb OpenAI launched two interfaces to GPT-3 at the same time: ChatGPT and InstructGPT. InstructGPT pretty much was what you describe, following natural language instructions without pretending to be human.
For whatever reason, and I think to OpenAI's surprise, ChatGPT became a global phenenon and heralded as the harbinger of the Singularity, while InstructGPT was forgotten.
As someone else pointed out, the AIs that take orders (Siri, Alexa, etc.) are female coded and met with "Meh. Kinda neat." The AIs that confidently spout bullshit are male coded and met with "We must give them hundreds of billions of dollars and put them in charge of everything!"
Make of that what you will.