The Summer Student
I’ve worked with Bard (now Gemini) and ChatGPT over the past little while and it reminds me of the sometimes difficult experience of working with a summer student. I once too was a summer student and remember being very motivated but ultimately completely incompetent at most tasks.
Sometimes, I still feel this way.
But, I was always willing to try. This is a how I feel about working with these LLMs at times.
For all the amazing feats of being able to riff and come up with slam poetry and meal ideas, LLMs can seem reluctant to exert the effort to overcome problems. I was reminded of working with summer co-op students.
Now, not all.
I’ve worked with some amazing co-op students, usually in their fifth or sixth co-op term where they’ve already had the equivalent of almost two years of work. And then I’ve worked with some first or second co-op term students. They were all lovely people with big hearts but boy, the effort needed to get some output was a lot. Not for lack of trying on their part — it was just that the figuring out how to convert initiative into productivity was lacking. You’d have to “re-prompt” them until they were able to get to the right solution.
This is the case with some LLMs today. ChatGPT gives me commentary when all I want to is a straightforward translation. It stops working after the equivalent of 5 PM (or the equivalent of when it needs to catch the bus) and then requires me to start over. There’s a lot of cutting and pasting, and reloading. “Hmm… let’s try a different approach,” I say.
While super smart today, the next generation of LLMs will do the work that needs to be done to do the real work. They’ll get out of their own way and re-prompt / correct as needed. That will be when we start moving back up the Trough of Disillusionment.
