I asked ChatGPT to write a poem on Explainable AI in John Keats' style and here is what I got. The irony of it is that I can't explain how ChatGPT arrived at this result. While there is much excitement about Generative AI, there is a growing sense that it would lead to black box models that we can’t see, control, or explain and therefore become a huge blind spot of trust, ethics, and accountability. Already the likes of Stackoverflow have banned ChatGPT posts. This is a warning call for our entire industry!
I read a comment on someone else's tries with ChatGPT (link below) and it kind of resonated with my thoughts - while at first glance it looks very good, a bit of scratching beyond the surface reveals it's actually not *that* good and misses a lot of the nuance. That said, indeed this would fool the vast majority of laymen who are but naive observers with neither the time nor knowledge of the specific matter the AI is claiming to reproduce.. we're in for some very interesting times going ahead. https://twitter.com/curiouswavefn/status/1599155339458641920?s=20&t=NaUEbDKXBcb2mdr81w2fPA
I asked ChatGPT for a short haiku about the DALEX package. The results are so frighteningly contextual.
Perhaps Stack Overflow banned because it’s a potential threat to its business? Saw in another post that one of the main uses of ChatGPT could be coders asking it questions when they’re stumped instead of going to Stack Overflow. I’m a novice programmer so I can’t comment on whether ChatGPT would be a better source for answers compared to fellow programmers, but maybe someone else here could?
Couldn't agree more Krishna Gade. We already have a lot of unexplainable AI / black box models floating around primarily in the D2C space...hope realistic expectations set in and the next ChatGPT release brings in explainability components built in. Enterprises need to very carefully assess real-world applicability and multiple scenario simulations (incl. robust anomaly detections on NLG output) before adopting ChatGPT at scale
It still lacks metaphors, definitely will fail turing test if the judges are expert poets. Nice try though.
Mitul Tiwari
1y
Agree. But that was a fun poem to read 😂
I liked the last part.. It's appreciating the power of explainable AI and i appreciate that.
ChatGPT has a tin ear, I'm afraid. It's pretty undergraduate stuff.
Pretty cool on the surface ... the deeper you dig, the scarier the implications!
Professor at St. Clair College
1yYeah, but you do have to admit that is a cool poem :)