AI is a tool (at least for now). And like many tools, when used proficiently, it can boost your productivity and therefore make you better, not redundant at your job.
> is it right 100% of the time, no.
No human teacher is going to be right 100% of the time either. I suppose ChatGPT is a little more overconfident / doesn’t have a sense of tone or anything to assess confidence; even still, though, given how overconfidence is incentivized in the workplace, it’s not that different in this regard from an average coworker.
100% agree. Its like having a coworker who you can constantly pester with technical questions. I really love how you can ask ChatGPT follow ups to earlier questions and it understand the context.
At some point, I expect companies to put their internal Wikis/documentation into LLMs so you can learn and onboard staff faster. You'd probably need a large amount of data to learn on, but I guess some of the large companies have that.
Basically, anyone who thinks that having a tutor-like experience with a robot is worth trading away correctness and consistency, is revealing that they care more about getting an answer in a form they can easily ingest than actually having any depth of understanding of the topic at hand.
And before someone makes the silly argument that "well your human tutor could be wrong too!", all of the training data that trained chatGPT was written by humans, and so the potential human error is still the same. ChatGPT is adding a new source of error that is unpredictable and based on dozens of anecdotes I have seen here and on Twitter, none of the enthusiasts are double checking it's output for bullshit
I don't have a human tutor available and to get one would represent a significant cost for me.
Slightly wrong and available is worth far more than always right and unavailable, especially at the early starting point where you're unsure of how much you want to invest into an endeavor.
You can still do this today without a financial cost, because the choice was never an either/or between paying for a human tutor and learning. And just because there are now nearly free and shitty tutors available, it doesn't mean learning should be done via these means / is effectively done via these means.
A recent example of this was some crap CSM my wife was working on that had piss poor documentation from the vendor. As in the vendor documentation was just flat wrong and would never work the way they wanted it implemented. We dumped a code snippet off in GPT4 and it came back with "This is wrong, do this instead" which worked.
The key point here is it solved a problem in a system that is rarely worked on. Going outside the vendor documentation to learn the python/js interaction the vendor obviously did not understand would have represented a huge investment over the payout of correcting the output of a single page.
Try asking ChatGPT about more complex stuff. Design patterns, computer theory. I think you'll be impressed at the results.
This is the strength of ChatGPT, for some (and possibly many) topics, it does a really good job. If it didn't, it would not be getting all the attention it currently commands.