Lawrence Phillips

[LW xpost] Unit eco­nomics of LLM APIs

• Lawrence PhillipsJul 12, 2024, 7:09 AM

  1 point

  0 ∶ 0

  in reply to: Aaron_Scher’s comment on: A break­down of OpenAI’s revenue

  There were humans in the loop, yes.

• Lawrence PhillipsJul 11, 2024, 9:33 PM

  1 point

  0 ∶ 0

  in reply to: Kei’s comment on: A break­down of OpenAI’s revenue

  The source for the $1B API revenue claim is given as “someone who viewed internal figures related to the business”.

  It’s not completely implausible, but the implications for OpenAI’s revenue growth curve would be a little surprising.

  We have fairly reliable numbers for ChatGPT Enterprise revenue (based on an official announcement of seats sold together with the price per seat quoted to someone who inquired) and ChatGPT Plus revenue (from e-receipt data) from the start of April; these sum to about $1.9B. It’s reasonable to add another $300M to this to account for other smaller sources – early ChatGPT Team revenue, Azure (which we did indeed ignore), custom models.

  So, with an extra $1B from the API on top of all that, we’d see only $200M revenue growth between the start of April and the middle of June, when it was announced as $3.4B – contrast with $1.2B between the start of January (December’s ARR was $2B) and March (estimated $3.2B).

• Lawrence PhillipsAug 31, 2022, 3:50 PM

  2 points

  0 ∶ 0

  in reply to: Peter Wildeford’s comment on: An­nual AGI Bench­mark­ing Event

  Thank you for the feedback. This is an important and valid concern. Similar concerns were raised on the discussion thread over at Metaculus, and we’ve replied with some thoughts there. It’s worth mentioning that I don’t think we should move forward with anything until we’ve carefully considered the consequences – probably using forecasting to help with this – and gotten feedback from several disinterested parties.

  I’ve thought a little more, at a very high level, about how an event like this might be designed in order to be beneficial overall, and written the idea up here.

An­nual AGI Bench­mark­ing Event