| ▲ | visarga 2 days ago | ||||||||||||||||
> algorithms just take other people's work, mix it so that any individual training input is unrecognizable and resell it back to them LLMs are huge and need special hardware to run. Cloud providers underprice even local hosting. Many providers offer free access. But why are you not talking about what the LLM user brings? They bring a unique task or problem to solve. They guide the model and channel it towards the goal. In the end they take the risk of using anything from the LLM. Context is what they bring, and consequence sink. | |||||||||||||||||
| ▲ | martin-t 2 days ago | parent | next [-] | ||||||||||||||||
Quantity matters. Imagine it took 10^12 hours to produce the training data, 10^6 hours to produce the training algorithm and 10^0 hours to write a bunch of prompts to get the model to generate a useful output. How should the reward be distributed among the people who performed the work? | |||||||||||||||||
| ▲ | lawlessone 2 days ago | parent | prev [-] | ||||||||||||||||
>But why are you not talking about what the LLM user brings? They bring a unique task or problem to solve. They guide the model and channel it towards the goal. In the end they take the risk of using anything from the LLM. I must remember next i'm shopping to demand the staff thank me when i ask them them where the eggs are. | |||||||||||||||||
| |||||||||||||||||