I've personally found AI to be a great help whenever I'm diving into a topic that I'm less familiar with. Recently I used it to help me prep for an interview as well. My partner uses it to help explain STEM concepts that she didn't cover in her schooling.
I do wonder how far away we are from an actual Young Lady's Illustrated Primer. Three years ago I'd say we were 50 years away. Now it feels more like 10.
I just don't know about this. I also find it answers great when I'm not familiar with a topic. However, when I am familiar with a topic I find all sorts of inconsistencies or wrong facts. I'm concerned the same inconsistencies are there in the topics I'm not familiar with, I just don't know enough about the subject to spot them.
I'm normally worried about that kind of Gell-Mann amnesia effect when I'm reading articles or watching YouTube videos that dive into a topic I'm unfamiliar with.
But when it comes to LLMs, I'm conscious of the fact that I'm asking deep, specific questions when it comes to programming and software, about things on the fringes that the LLM can't have read much about. In contrast, with things outside my area of expertise, I'm asking very superficial question that any practitioner of the field could answer, and so likely has many many sources for the LLM to pull information from.
With fields outside my area of expertise, I'm most often asking questions that are comparable to "what is a variable" or at most "how does A* search work". That gives me confidence that the quality of answers is likely to be much better with these questions.
I've noticed the same thing, specifically when asking about programming topics. It reminds me of the Gell-Mann Amnesia effect, but at least you're aware of the inaccuracies.
The same for me, I love it for this sort of thing. I can bounce ideas off of it and it'll give me a solid response without getting tired of my questioning. And it'll explain in detail why I'm wrong. I really can't express how useful this is for my style of learning - I like to take things apart and figure out how they go back together.
This is exactly it, this sort of infinitely patient tutor dialog interaction is absolutely perfect for my style of learning and I can't help but be a little bit sad that I didn't have something like that available during school. That being said, I'm beyond thrilled to have it available now, makes it easy to never stop learning.
Today I asked ChatGPT to implement an oauth 2.0 token flow for me in bash. I tweaked a few parameters so it’d feed directly from a settings file in the folder I’m running it from. Then I realized we needed it in powershell, I pasted the tweaked script in and said “write this in powershell” and the script just WORKED. It was great. I don’t care about the oauth 2 workflow it’s well documented and implemented in our code in 50 places, I just wanted the script so I can integrate it into some automated testing. It would have taken me probably an hour of messing up syntax in bash and then figuring equivalents in powershell to finish up the script.
My coworkers were absolutely thrilled at my turnaround time. I was thrilled I didn’t have to do this boring task. I saw this as an absolute win.
I do wonder how far away we are from an actual Young Lady's Illustrated Primer. Three years ago I'd say we were 50 years away. Now it feels more like 10.