“Ph.D. level intelligence” is an oxymoron, because it’s not intelligence that defines a Ph.D. LLMs lack curiosity, enthusiasm, or interest, nor can they be responsible for their own research. LLMs can’t even ensure the truthfulness of their own generated text.
The AI industry is being driven by slogans designed to make people believe it's revolutionary rather than just somewhat useful. The investment is off the charts, and they are worried about paying the bills.
Particularly since scientists and engineers don't have a an agreed defintion of what "intelligence" actually is.
I've followed AI and cognitive science since 1990. (When Japan announced that it would have a fully-functional AI by 2000. 😆😆) We still have no significant agreement on what it is.
For me, AI isn't even an intelligence. It's a voice driven Internet appliance augmented with a profile of human contextual expectations. And intelligence would (IMO) would have self-motivation.
And we really REALLY don't want that to happen!
I'm sitting here giggling because an LLM is also a legal (post-JD) scholar. And yet given the legal profession's slavish devotion to precedent, the statement may very well still apply!
If you read past the headlines, this quote is taken specifically by Microsoft as well as OpenAi's CTOs who suggested that the next generation of chatgpt will be capable of passing PhD level exams.
Murati echoed that this is also task specific. So no, no one's suggesting any real intelligence here
But there’s no such thing as a PhD-level exam: a PhD is a research degree which specifically requires you to make a novel contribution to the field in your thesis. You can’t obtain a PhD simply by regurgitating existing knowledge- and that’s a bit of a problem for an LLM.
If you think what I said showed that I've forgotten, I suggest you re-read what I wrote - and as for assuming I don't know what I'm talking about, I'll let you get on with that :).
I look forward to an LLM taking a viva and, in response to being asked "Did you consider X, Y and Z?", telling the examiners to eat rocks.
Might well be a more honest rendition of the candidate's internal monologue than what the candidate would actually say, of course.
Yeah, I know the meetings:
CTO: So… can we say this? [marketing pitch]
Developers: [long, detailed explanation of why we cannot in fact say that]
CTO: So that works then? Great! [gets up, hurries out]