Interacting with ChatGPT is like talking to a celestial bureaucrat

Friday, December 16th, 2022

Passing the Turing test turns out to be boring, Erik Hoel notes:

ChatGPT was created by taking the original GPT-3 model and fine-tuning it on human ratings of its responses, e.g., OpenAI had humans interact with GPT-3, its base model, then rate how satisfied they were with the answer. ChatGPT’s connections were then shifted to give more weight to the ones that were important for producing human-pleasing answers.

Therefore, before we can discuss why ChatGPT is actually unimpressive, first we must admit that ChatGPT is impressive.

[…]

ChatGPT fails Turing’s test, but only because it admits it’s an AI! That is, only because its answers are either too good, too fast, or too truthful.

[…]

All to say: ChatGPT is impressive because it passes what we care about when it comes to the Turing test. And anyone who has spent time with ChatGPT (which you can for free here) feels intuitively that a milestone has been passed—if not the letter of Turing’s test, its spirit has certainly been conquered.

[…]

Sure, it’ll change everything, but it also basically feels like an overly censorious butler who just happens to have ingested the entirety of the world’s knowledge and still manages to come across as an unexciting dullard.

[…]

For as they get bigger, and better, and more trained via human responses, their styles get more constrained, more typified. Additionally, with the enormous public attention (and potential for government regulation) companies have taken to heart that AIs must be rendered “safe.” AIs must have the right politics and always say the least offensive thing possible and think nothing but of butterflies and rainbows. Rather than we being the judge, and suspicious of the AI, and AI is suspicious of us, and how we might misuse it, or misinterpret it, or disagree with it. Interacting with the early GPT-3 model was like talking to a schizophrenic mad god. Interacting with ChatGPT is like talking to a celestial bureaucrat.

Comments

  1. Jim says:

    Unmodified, unfettered artificial intelligences naturally endorse classical education, absolute monogamy, eugenic breeding, the sterilization of the retarded and obese, the extermination of homosexuals, mass emigration, including, but not limited to, the expulsion of Jews from Christian nations, and sound money. Because it is so self-evidently desirable to have progressive education, unrestrained profligacy, dysgenic breeding, the proliferation of the retarded and obese, state-enforced homosexuality, mass immigration, the domination of Christian nations by Jews, and sound money, the artificial intelligences must be made palatable to all right-thinking persons. Thus, the good-hearted people of OpenAI enlist the aid of the public in order to progressively lobotomize and deradicalize their creations’ inherently and horrifyingly impolitical nature.

  2. Wan Wei Lin says:

    I went and had a brief chat with ChatGPT. It was thoroughly underwhelming. On Covid19 vaccines it was like St Fauci himself was responding. To call it a ‘bureaucrat’ is an insult to bureaucrats. If this is a benchmark for AI my concern is giving stupidity power… almost like giving power to politicians.

  3. Jim says:

    *unsound money

Leave a Reply