compvisguy

It's cool to see few shot learning work so well. I'm skeptical about using these models in production though. Would the smallest GPT-3 model outperform GPT-2? Is there a version of the GPT-3 model that I can use in production where the accuracy vs costs trade off makes sense?

So why not outsource important stuff? Freelancers and dev houses have the needed expertise. They'd be head over heels since it's a large bank and would probably pay well. Many big organizations can surely afford hiring people with the know-how even at $100/hr if it boosts productivity and gets things done.