Hello Bob,
Am 09.01.2025 um 00:24 schrieb Bob Cagle <bcagle@xxxxxxxxxxx>:
As an aside, I'm also curious how your users are utilizing these tools in their jobs outside of the ERP systems? I see a lot of promise for assisted coding, creating digital art, summarizing YouTube videos, and analyzing data to a certain degree. But what more useful things can these things do that we can help our users utilize them to their full potential?
Many people apparently use ML as a mere toy, to generate funny images, internet memes or play with it to see how "intelligent" it is.
Having started testing ChatGPT only fairly recently, my experience is:
- Garbage in, garbage out. It has been trained from stuff on the internet and the internet is full of sh*t.
- You can't just trust what it comes up with. Example: I've asked it what MULIC and FULIC means and it made up "multi-volume interactive cartridge", and "full interactive cartridge". Always do secondary checking. Depending on what you want to achieve, the extra time asking ChatGPT in the first place might be completely wasted.
- The output heavily depends on topic. The more stuff the net has about something, the more the generated output might be correctly derived. For now and the foreseeable future, checking is mandatory.
Because of the mandatory checking part, I largely question ML's real world value. Blindly trust a YouTube video summary until you find it's subtly wrong? Cursing because the proposed code snipped doesn't work as copy-pasted? Wrong conclusions from data analyzation leading to losing money? Six fingered hands?
I assert, the technology is still in its infancy and has nevertheless been released to the public, fostering the next hype in the IT industry, always seeking new opportunities to feed their shareholders. Media fed the hype with untenable promises, creating an almost perfect feedback loop with the moneymakers. And people sit in between, are fascinated about how human-like the output seems and reflect their own desires into "AI", including those who firmly believe they're chatting to an intelligent digital animate being.
I refuse to call what we currently have AI, because it's very artificial, and all but intelligent. It's ML, and to me, real world use cases are fairly limited because you just can't trust the output for factual correctness. ML is — somewhat simplified — a program having been written by a program being fed by inconceivably large amounts of "example data".
Yet, there are actually helpful use cases. Such as code snippet proposals as a starting point. That might help for not starting from 0. But then, the same can be found on the net with ordinary searches. It must, because that's where ChatGPT got its input data from.
:wq! PoC
midrange.com
