AI: Will someone open-source a $100M model by 2025?
➕
Plus
17
Ṁ479
Jan 1
29%
chance

Weights.

Get Ṁ1,000 play money
Sort by:

IIRC, stable diffusion was ~$0.5M, GPT3 maximum cost estimate was ~$5M, so $100M is a big jump to make in 3 years. Unless you count costs of failed training runs, but even then, $100M is a lot, especially as hardware gets more optimised for AI.

How do we evaluate $100 million for in the model?

Is it just compute cost or also labour cost?

How will we calculate the cost of the compute if it’s community trained through grid computing or equivalent “compute donation” mechanism instead of direct purchase?