This thread is already full of people throwing in their cheap takes by reading the headline or not even that. Listen to his answers. He's not making controversial claims, excepting the one that LLMs teach us nothing about language. He says: - ChatGPT makes it easier to cheat. This is obviously true. - ChatGPT is a method of avoiding learning. In the context of the interview he's referring to its use not as an informa…
Or in other words, it is great at being a confident sophist and its use-case(s) are limited to almost nothing of serious use requiring explainability or trustworthiness; especially for legal, medical and financial applications or even search engines. If you cannot trust the output of an AI and it keeps hallucinating its answers, then what is the point?
If OpenAI was being 'responsible' or following 'responsible AI' guidelines, maybe they should be writing the detectors first before releasing ChatGPT, GPT-3 into the wild?
This AI hype cycle is no different to what happened with GPT-3 and it fizzled out just like how Clubhouse did. The only way is open source AI models.