A 42-minute CBS interview with Geoffrey Hinton at the Vector Institute in 2023, covering the history of neural networks and the present and future of AI. He discusses AGI timelines and whether language models truly understand.
Why Hacker News points people here: Recommended as a more nuanced look at Hinton's views than headlines, especially his argument that LLMs are more than auto-complete.
“found the discussion much more nuanced than today's headlines.”HN comment, 2023
“Geoffrey Hinton recently gave a very interesting interview and he specifically wanted to address the "auto-complete" topic”
Hinton addresses the auto-complete characterization of LLMs.
Full comment
If you (or anyone else is) interested in the topic, I'd highly recommend giving some of these a look: Bubeck, Sébastien, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, et al. “Sparks of Artificial General Intelligence: Early Experiments with GPT-4.” arXiv, March 27, 2023. link . or watching Sebastien Bubeck's recent talk he gave describing what GPT-4 can do that previous LLMs couldn't: link Geoffrey Hinton recently gave a very interesting interview and he specifically wanted to address the "auto-complete" topic: this video Here's another way that Ilya Sutskever recently described it (comparing GPT 4 to 3): link I'd also recommend this recent Sam Bowman article that does a goood job reviewing some of the surprising recent developments/properties of the current crop of LLMs that's pretty fascinating: Bowman, Samuel R. “Eight Things to Know about Large Language Models.” arXiv, April 2, 2023. link .
RecommendstpowellMay 2023
“found the discussion much more nuanced than today's headlines.”
More nuanced full interview than the headlines suggest.
Full comment
Yesterday, I randomly watched his full interview from a month ago with CBS Morning, and found the discussion much more nuanced than today's headlines. this video The next video in my recommendations was more dire, but equally as interesting: link
RecommendsstareatgoatsMay 2023
“Don't take my word for it, listen to Geoffrey Hinton explain it instead”
Hinton explains LLMs have a real form of understanding.
Full comment
No, we're past that point. it's no longer the most useful way to describe these things, we need to understand that they already have some sort of "understanding" which is very similar if not equal to what we understand by understanding. Don't take my word for it, listen to Geoffrey Hinton explain it instead: this video
Cites itkarmasimidaMar 2023
“Hinton said he didn’t believe it before ChatGPT came out, that AGI is possible in 20 years.”
Hinton's own words on AGI timeline.
Full comment
this video Hinton said he didn’t believe it before ChatGPT came out, that AGI is possible in 20 years. It is simple to just checkout peoples’ own words, otherwise you are hallucinating the same way as those LLMs do
Cites itHerval_freireApr 2023Full comment
>Well, that's the capital equipment case then. Is it really a general intelligence if you don't let it be general? It can't handle every possible scenario for you that way. Humans are most useful to other humans as happy slaves. Autonomy only is good for humans as individuals. If we create useful agi of course we will bend it to our will. There is no incentive to make AI as a separate competitor other than an experiment. Either way this is hypothetical. >If your theory is that someone will program an AGI for the specific goal of making you unemployed, sure that could happen. But it sounds more like warfare than economics to me at that point, so I can't comment cause I don't know too much about that. No it's cost saving. I don't want to employ someone who I have to pay. This is capitalism. I save money by using AI. >Where's the data? That was pure theory there, I didn't make any empirical claims. You don't need to cite your sources. It's obvious you're getting your conclusions from data. You use the words "in practice". I agree with you on how the data plays out. >It's important to do this because approximately all AGI doom on the internet comes from Big Yud, an internet cult founder who thinks you can predict the future by thinking about it really hard and doing probability theory in your head without needing to check any assumptions. Which is why he thinks AGI is going to enslave him, because it's on a computer so it can do math exponentially faster, never mind that it can't
