They may panic, but they shouldn't. They can quickly pivot. GPT programs can be used off the shelf, but they can also use custom training. Every large org has a huge internal set of documents, plus a large external set of documents relevant to its work (research articles, media articles, domain relevant rules and regulations). They can train a GPT bot to their particular codebase. And that is now. Soon (I'd give it at most one year), we'll be able to train GPT bots to videos.
Yeah this thread has been the motivation for me to sign up on the wait list and cost out what it would take to try fine-tuning their older models on our data. There's still plenty of work out there when it comes to building a solution to a problem.
100%. Anybody with experience in distributed systems, networking, or SRE knows the plumbing can be as challenging as the “big idea”. Training these models is a plumbing job. And that’s actually really hard to pull off.
All this training does not happen by itself.