4 ms·

Has anyone articulated a theory for what the profit model might actually look like for the big LLMs? In the .com era, the valuations were crazy, but people gene

Has anyone articulated a theory for what the profit model might actually look like for the big LLMs? In the .com era, the valuations were crazy, but people generally understood how you could eventually make money on all of this stuff. If models are requiring larger and larger infrastructure buildouts, does anyone have a clear sense of what users will have to pay in order to make the businesses profitable?

sunir 1y ago
The dot.com had no idea. They talked about eyeballs.
- SoftTalker 1y ago
  I think they had an idea that it would be fees per transaction or a subscription model. They just had no customers.
  - sunir 1y ago
    They all thought it would be advertising or ecommerce. Subscriptions weren't a cultural phenomenon yet. That was a decade later.
- bparsons 1y ago
  They knew you could sell stuff on the internet. The payment processors were just clunky. It was fairly obvious at the time how it would all work. I think the surprise was the degree to which ad revenue would eat the world. Maybe it will be the same this time.
micromacrofoot 1y ago
no one can predict the future, they were guessing then and they're guessing now guesses in the past look better because we tend to pay more attention to the correct ones
dgs_sgd 1y ago
With the direction of OpenAI, hyper personalized ads inserted directly into chat and their app experiences could be a path. Not saying it will work, but they’re definitely exploring it.
HarHarVeryFunny 1y ago
It seems that each model generation, and each generation of GPUs, has a limited shelf life, so you need to make more money from each generation of models/GPUs than it cost you to build/buy them. User pricing to make a profit depends on inference volume. You need to make back a fixed dollar amount before breaking even, which could come from high volume and low prices, or lower volume and high prices. The trouble is that everything is changing so fast that any kind of forecasting is extremely error prone, especially when one forecast builds on another. First you need to guess how much more capable the models are going to get (at things where people will pay more for better performance), in what time frame, then guess what level of demand (inference volume) will exist with that level of capability... The LLM developers like OpenAI and Anthropic like to tout things like Math Olympiad and Competitive Programming results as signs of progress, but there is no guarantee that they will be equally successful in applying RL to more general areas of commercial value where RL rewards are harder to define. These companies also like to talk about "scaling laws" is if there was some inevitability about investing more money & compute and getting better results, but this only works until it does not, and they replace one broken "law" with another. Right now it's all about scaling of RL-training and test time compute, but how long will that last, and what type of problems will benefit? The profit model here seems a bit like the Drake equation for calculating the probability of other intelligent life in our galaxy... it may be possible to define the equation, but the outcome depends on having the right values for all the variables, which are largely unknown.

sunir 1y ago

The dot.com had no idea. They talked about eyeballs.

SoftTalker 1y ago
I think they had an idea that it would be fees per transaction or a subscription model. They just had no customers.
- sunir 1y ago
  They all thought it would be advertising or ecommerce. Subscriptions weren't a cultural phenomenon yet. That was a decade later.
bparsons 1y ago
They knew you could sell stuff on the internet. The payment processors were just clunky. It was fairly obvious at the time how it would all work. I think the surprise was the degree to which ad revenue would eat the world. Maybe it will be the same this time.

micromacrofoot 1y ago

no one can predict the future, they were guessing then and they're guessing now guesses in the past look better because we tend to pay more attention to the correct ones

dgs_sgd 1y ago

With the direction of OpenAI, hyper personalized ads inserted directly into chat and their app experiences could be a path. Not saying it will work, but they’re definitely exploring it.

HarHarVeryFunny 1y ago

It seems that each model generation, and each generation of GPUs, has a limited shelf life, so you need to make more money from each generation of models/GPUs than it cost you to build/buy them. User pricing to make a profit depends on inference volume. You need to make back a fixed dollar amount before breaking even, which could come from high volume and low prices, or lower volume and high prices. The trouble is that everything is changing so fast that any kind of forecasting is extremely error prone, especially when one forecast builds on another. First you need to guess how much more capable the models are going to get (at things where people will pay more for better performance), in what time frame, then guess what level of demand (inference volume) will exist with that level of capability... The LLM developers like OpenAI and Anthropic like to tout things like Math Olympiad and Competitive Programming results as signs of progress, but there is no guarantee that they will be equally successful in applying RL to more general areas of commercial value where RL rewards are harder to define. These companies also like to talk about "scaling laws" is if there was some inevitability about investing more money & compute and getting better results, but this only works until it does not, and they replace one broken "law" with another. Right now it's all about scaling of RL-training and test time compute, but how long will that last, and what type of problems will benefit? The profit model here seems a bit like the Drake equation for calculating the probability of other intelligent life in our galaxy... it may be possible to define the equation, but the outcome depends on having the right values for all the variables, which are largely unknown.