Selling embedding is genius, it amortizes the cost of training huge models over thousands of use cases. Soon it will become very profitable to x2 or x3 the number of parameters, as value created by better embedding / completion will cover the cost, this process repeating at each new model relase

ser, what is the smallest unit embeddings and how do we measure the quality? or in general how to price it?

building degencast.wtf /degencast, ex-synfutures, binance, bitmain

Depends on your use case. If you build a recommendation engine for ex, the value of your embedding can be calculated by the ability to predict a user's likes using them. Not sure what you mean by smallest unit embeddings ?

as in how do you sell and how do you price it assuming there is an active market for it

So you sell just like openai does, charging per token, and you just price it based on the cost of generating it + your margin