Uploaded March 2024 | Updated September 2026, 2 weeks ago
Check out my essays: aisc.substack.com
OR book me to talk: calendly.com/amirfzpr
OR subscribe to our event calendar: https://lu.ma/aisc-llm-school
AF: You started your talk by pointing out that example that happened a few years ago around sentience. So I can't help myself, but to go there. Language is a poor representation of intelligence. It is one of the best tools we have come up with as sapiens, and there's a lot of conversation going on around sentience and consciousness and general intelligence when we are seeing that LLMs can parrot humans in terms of language.
I'm just going to use that as a prompt and take it where you want from here.
JA: Two things I love to say in reaction to that. One is that software is already so much better than humans in a lot of different things, but we just keep moving the goalposts.
Then the other thing is that there's a lot of research on just the language centers of the human brain. There are these conditions where that language center can be eradicated and a person can lose their language ability, but they remain able to grow, operate in the world, go out and buy things and just operate a household completely.
We don't yet have an equivalence of a language model being a brain, and I think we're far from it. But we do see these sparks of "yes, this is smarter than anything we've had before", and it will continue to get smarter, and that is exciting.
Remains to be a philosophy, but we still don't understand the human brain enough to say anything more.
This is another rat hole that I found very exciting, which is to break down the communication barriers between humans and other species. There's a lot of work in using AI to decode the language of whales and dolphins and elephants. As you think about intelligence, that, to me, is extremely exciting work and application of AI.
AF: What is exciting for me is that, while it doesn't capture the full picture, it is a poor representation, but it's a means to an end, which is connecting a lot of different expert systems and expert minds to do a collectively interesting job together.
Check out my essays: aisc.substack.com
OR book me to talk: calendly.com/amirfzpr
OR subscribe to our event calendar: https://lu.ma/aisc-llm-school
AF: You started your talk by pointing out that example that happened a few years ago around sentience. So I can't help myself, but to go there. Language is a poor representation of intelligence. It is one of the best tools we have come up with as sapiens, and there's a lot of conversation going on around sentience and consciousness and general intelligence when we are seeing that LLMs can parrot humans in terms of language.
I'm just going to use that as a prompt and take it where you want from here.
JA: Two things I love to say in reaction to that. One is that software is already so much better than humans in a lot of different things, but we just keep moving the goalposts.
Then the other thing is that there's a lot of research on just the language centers of the human brain. There are these conditions where that language center can be eradicated and a person can lose their language ability, but they remain able to grow, operate in the world, go out and buy things and just operate a household completely.
We don't yet have an equivalence of a language model being a brain, and I think we're far from it. But we do see these sparks of "yes, this is smarter than anything we've had before", and it will continue to get smarter, and that is exciting.
Remains to be a philosophy, but we still don't understand the human brain enough to say anything more.
This is another rat hole that I found very exciting, which is to break down the communication barriers between humans and other species. There's a lot of work in using AI to decode the language of whales and dolphins and elephants. As you think about intelligence, that, to me, is extremely exciting work and application of AI.
AF: What is exciting for me is that, while it doesn't capture the full picture, it is a poor representation, but it's a means to an end, which is connecting a lot of different expert systems and expert minds to do a collectively interesting job together.
