Chamath

If like the level one customer support person can't answer the question, they kick it up to level two. So... There's a place for... LLMs to start in customer support, which is replacing all the level one and then working their way up the chain to level two as they get better and better.

If like the level one customer support person can't answer the question, they kick it up to level two. So... There's a place for... LLMs to start in customer support, which is replacing all the level one and then working their way up the chain to level two as they get better and better.

And so what I'm saying is that the level of accuracy now, especially with the new PhD level reasoning models is good enough. We don't need to wait for like some perfect LLM model. And I think this is why this is going to be a big, big disruption. Millions of people potentially are going to have their jobs disrupted or at least transformed.

And so what I'm saying is that the level of accuracy now, especially with the new PhD level reasoning models is good enough. We don't need to wait for like some perfect LLM model. And I think this is why this is going to be a big, big disruption. Millions of people potentially are going to have their jobs disrupted or at least transformed.

Well, it could be the end of the entire career as well, Chamath, if you were to look at this four by four sort of quadrant chart that Sachs is describing, which is the cost of an era, you know, and the actual complexity of the job, perhaps, or the cost of the job. How do you look at this? I know you're working on software that kind of does this with your startup as well.

Well, it could be the end of the entire career as well, Chamath, if you were to look at this four by four sort of quadrant chart that Sachs is describing, which is the cost of an era, you know, and the actual complexity of the job, perhaps, or the cost of the job. How do you look at this? I know you're working on software that kind of does this with your startup as well.

Yeah.

Yeah.

It's really... Have you guys worked with the O1 preview yet? I just literally have been using this new reasoning engine that... OpenAI released, and it is extraordinary. And it's kind of thinking about the next three or four prompts you would do. And I literally just got this while we're on the show. I've hit the limit for my paid account because this thing is so intense on compute, I guess.

It's really... Have you guys worked with the O1 preview yet? I just literally have been using this new reasoning engine that... OpenAI released, and it is extraordinary. And it's kind of thinking about the next three or four prompts you would do. And I literally just got this while we're on the show. I've hit the limit for my paid account because this thing is so intense on compute, I guess.

Right.

Right.

One of the reasons why I'm bullish on this customer support use case is because there's a very large data set to train on. You've got all of the product documentation that companies have already created. You've got all of the previous email support. And calls. And calls, yeah. The calls have been recorded, so you can now train the AI on that. So there's a very large...

One of the reasons why I'm bullish on this customer support use case is because there's a very large data set to train on. You've got all of the product documentation that companies have already created. You've got all of the previous email support. And calls. And calls, yeah. The calls have been recorded, so you can now train the AI on that. So there's a very large...

body of data to train the AI model on. And it's not necessarily the most proprietary. It's not like dealing with people's medical records or even confidential legal documents, something like that. So the data is readily available. And then the foundation models are getting really good.

body of data to train the AI model on. And it's not necessarily the most proprietary. It's not like dealing with people's medical records or even confidential legal documents, something like that. So the data is readily available. And then the foundation models are getting really good.

I think there's a big question here about value capture, which is there's a number of startups now that are becoming very highly valued that are chasing this disruption, this sort of customer support agent disruption. And they're getting into very high valuations, even unicorn valuations already.

I think there's a big question here about value capture, which is there's a number of startups now that are becoming very highly valued that are chasing this disruption, this sort of customer support agent disruption. And they're getting into very high valuations, even unicorn valuations already.

And the question is, well, wait, if the foundation models are advancing at such a rate, like a year from now, why couldn't a developer, just a startup of a few guys, take next year's model and train it and then commoditize the

And the question is, well, wait, if the foundation models are advancing at such a rate, like a year from now, why couldn't a developer, just a startup of a few guys, take next year's model and train it and then commoditize the