Apr 28 • General discussion
Can AI Get Better and Cheaper Faster Than Expected?
When Brimma started doing document AI with OpenAI tech almost a year ago, the cost per document was close to $1. There is not much of an economic model to use something that expensive. Fortunately, a year later, the cost is a small fraction of that. There are already "laws" people have proposed on how fast AI capability will expand and cost will decrease (e.g. "Hyper Moore's Law")
The question that occurs to me with this article is whether these "laws" are too conservative. That is, will AI get better and cheaper even faster?
Why do I think that may be the case? In the article, there are two elements that will be key:
First, DeepSeek is putting head-on pressure on all other models to deliver at a fraction of the cost. 97% cheaper is more than throwing down the gauntlet.
Second, DeepSeek's tests were run on Huawei's chips...which is a throw-down to every using Nvidia or their own custom chips.
I'm not betting against OpenAI, Gemini, or Nvidia, but this arms race is unbelievable!
0
2 comments
Can AI Get Better and Cheaper Faster Than Expected?
skool.com/lending-ai-free-community-2711
A FREE community to exchange ideas and share resources that allow you to stay abreast of what AI progress is tangibly being made in the lending domain
Powered by