Post by @seb
Well, I finally had an "AI in the workplace" meeting...
As an external contractor, I'm privileged that my opinion matters to them, and I _think_ I've managed to convince them that having an LLM pretend to be human to say nice things to the staff members and customers is a bad idea.
I was also able to demonstrate the privacy/reputation/security issues with some things they have already done, i.e. creating "apps" (static HTML and js webpages) and uploading them to varying services (vercel, nelify, etc.) Is a baaaaad idea.
And after extracting the "what problem is it you're trying to solve" question, another couple of people and myself were able to say "why haven't you asked for that before? It would just be a simple report!"
They may still have "AI" in their future in some form, but I think I've managed to cease the idiocy for now.
"It's only going to get better" was said though, to which I responded "it fundamentally cannot"... And then described the economical, environmental and general exponential growth problem/power requirement of having hundreds of billions of tokens in a model to actually get better.
Did I do well?
I think you did well. It's hard to deal with this in the workplace. My $workplace hands out licences like candy and a lot of people are drinking the Kool Aid.
1 like
@seb@fedi.dlny.eu
As much as I hate to admit it, there is a positive to a business handing out licences - at least if staff are using AI as a business it's within an enterprise version (different T&Cs) instead of everyone having some random public chatbot - they absolutely should not be uploading sensitive information to a public chatbot! (They are though...yikes!)
It's a single positive of course, it's negatives everywhere else after that.
Indeed, different T&Cs are an advantage, but still. Where does the data go? And regular users go for ease of use. If security awareness is any indication, it doesn't bode well.
The only acceptable use IMHO are local (to the data owner) instances that use ethically trained models.
@seb@fedi.dlny.eu
With you 100% there.
That did come up in the meeting actually - I asked "do you want your chatbot to just provide generic information it found on the internet that $business may not necessarily agree with, or a chatbot trained directly on your own data and processes so it's only providing answers from that data?"
The answer is, of course, what any caring business would say - I want it to say what I would say.
@paul@paulwilde.uk
Exactly, context matters. Who needs the whole internet as context and be average in the process? :)
Maybe new jobs will emerge: Context Engineer - natural person tasked with the responsibility of providing the right context/training data for the company's local LLM in an ethical way, taking into account the:
* the company's values and standards
* the company's unique selling points
* laws and regulations for all the regions the company is active in
Just pondering...