LLM costs are following a much more rapid exponential downward cost curve than even moore’s law (cut in 1/2 every 18 months).
According to some research I just did:
Over the 3-year period from 2021-2024, LLM cost efficiency improved by 1000x while Moore’s Law delivered a 4x improvement, making LLM progress 250 times faster than the traditional semiconductor scaling rate.
For example, o3 was released as flagship model cheaper than it’s predecessor, and 3 months later, they cut the price by 80% on the release of o3-pro.
To me this has some very important implications for startup operators:
- you can now do 1000x more work, without raising capital
- you can do hyper-personalization for customers and outreach, at scale
- you have to build for “where the puck is going” because today’s needs will be solved by the time you build for them
Seriously want some feedback here - am I just buying all the hype or is there really some truth to this? How are you changing how you operate as a result of how fast and cheap AI is?