3 ms·

This is a contrarian opinion and probably wrong. If you think current AI models are not economically viable, you need to reexamine your world view

matthewdgreen 3y ago
Could you elaborate a bit?
- elorant 3y ago
  Every company out there, from the smallest to the Fortune top-500 ones, could utilize a language model. You can use it for pretty much everything: marketing, sales, data analysis, customer support, market research, product development— you name it. Given a few years, it will become clearer what it can be used for, and their revenues could reach dozens of billions. Meanwhile, if they can manufacture their own chips, which is evidently their goal considering the significant investment, they could leave their competitors years behind.
  - mattlondon 3y ago
    Most of the other players have their own hardware too though - Google, Amazon, MS at least, and who knows what apple is up to (if anything ). Are OpenAIs chips going to be years ahead of those guys, or even Nvidia? I am sceptical.
    - elorant 3y ago
      They don't need to be years ahead. Availability and ideally a lower cost is the goal.
    - fxtentacle 3y ago
      Most AI using companies want something like a 3090. Performance and price are both great for enterprise use. But due to NVIDIAs EULA, you're not allowed to use them in the datacenter and that forces you to buy A100s at 20x the price. So just getting close to NVIDIAs retail price and performance, but without that EULA, would already be massively valuable.
  - matthewdgreen 3y ago
    The point I made in the original post was that having the best language model today doesn’t produce substantial revenue. Rather the bet here is that they will continue to have the best language model (and chips etc.) when we reach the commercialization period you refer to when you say “given a few years.” That’s a pretty long term bet on the idea that OpenAI will retain its current advantages and that OpenAI’s best model will be the only one that can satisfy those use-cases well and that OpenAI will be better at selling to the specific businesses that need the tech. Maybe all those things will remain true, who knows.
    - elorant 3y ago
      Well they have Microsoft's back who knows better than anyone how to sell to the enterprise.

matthewdgreen 3y ago

Could you elaborate a bit?

elorant 3y ago
Every company out there, from the smallest to the Fortune top-500 ones, could utilize a language model. You can use it for pretty much everything: marketing, sales, data analysis, customer support, market research, product development— you name it. Given a few years, it will become clearer what it can be used for, and their revenues could reach dozens of billions. Meanwhile, if they can manufacture their own chips, which is evidently their goal considering the significant investment, they could leave their competitors years behind.
- mattlondon 3y ago
  Most of the other players have their own hardware too though - Google, Amazon, MS at least, and who knows what apple is up to (if anything ). Are OpenAIs chips going to be years ahead of those guys, or even Nvidia? I am sceptical.
  - elorant 3y ago
    They don't need to be years ahead. Availability and ideally a lower cost is the goal.
  - fxtentacle 3y ago
    Most AI using companies want something like a 3090. Performance and price are both great for enterprise use. But due to NVIDIAs EULA, you're not allowed to use them in the datacenter and that forces you to buy A100s at 20x the price. So just getting close to NVIDIAs retail price and performance, but without that EULA, would already be massively valuable.
- matthewdgreen 3y ago
  The point I made in the original post was that having the best language model today doesn’t produce substantial revenue. Rather the bet here is that they will continue to have the best language model (and chips etc.) when we reach the commercialization period you refer to when you say “given a few years.” That’s a pretty long term bet on the idea that OpenAI will retain its current advantages and that OpenAI’s best model will be the only one that can satisfy those use-cases well and that OpenAI will be better at selling to the specific businesses that need the tech. Maybe all those things will remain true, who knows.
  - elorant 3y ago
    Well they have Microsoft's back who knows better than anyone how to sell to the enterprise.