Accountability, or lack thereof
Yes, LLM's aren't perfect. That's not the gotcha you think it is.
Cross-posted from AI Marketing for Storytellers.
So, I spent about 36 of my waking hours this past week lowkey worried that I killed someone.
I didn’t, actually. The train-wreck was wrecking before I came on, and d/t circumstances that I can’t mention here, there was never any way to un-wreck it.
But — so, I loved The Pitt, right? It was the first medical drama that took nursing seriously and made me feel super fucking seen, especially after covid, and there was a part in it, after a mass casuality, where someone asks a nurse how she can make it through the death and destruction, and my husband shouted out from his spot beside me on the couch, “She comes home and yells at her spouse!”
Which…is true. Heh.
Basically, after a shitty shift, I come home, I sit in my car staring at my phone for an additional 15 mins and/or until my husband texts me, “Wanna eat?” or “Coming in?” ‘cause he can see me from his office window, and then I come in and I don’t talk about anything, until ten at night when we’re about ready to go to sleep, and by then I’ve processed my grief into digestible chunks and I’m all, “DO YOU WANT TO KNOW HOW TODAY WAS?”
And he’s all, “Please, take your ambien. Please.” (because he has to work in the morning too, ofc) — so I do, benzos being my one vice/drug of choice — if I cannot take ambien to sleep after having been a goddamned covid ICU nurse, I will cut somebody about it, you do not have the nightmares that I have, we are not the same — and I do and I pass out and by the time I wake up, usually I’ve put it into a box and throw it into the box-hole in my soul where all the other shit I’ve had to go through and/or take goes, because that’s as emotionally healthy as you can get when you’re still involved in an abusive relationship with your chosen profession.
So — I think about that, a lot. And how I relate to that.
Because my day job is all about acknowledging that there’s a chance I’ve messed up.
I cannot afford to have pride about errors when someone else’s life is on the line.
I have pride in my work, yes, certainly, 100%.
But I can’t be all, “Oh, I’ve got this covered,” when someone’s on five different blood pressure medications awhile we dogpaddle and wait to see if Jesus is going to be on our side.
And I think, in a certain way, that’s made me way more able to use an LLM than the normal person/creative type.
You see — I have to acknowledge, every single ding-dong day I’m here, that, while being incredibly good at my job and competent that I:
A) Might fuck up terribly, despite my best intentions, and accidentally hurt someone and bear the scar of that on my soul for the rest of my mortal days and
B) Might do my best fucking job ever and it still not fucking matter because giving healthcare is not a meritocracy and sometimes you just don’t get to win.
It’s a really humbling thing, over, and over, and over again.
Endlessly so.
It’s just how it is.
And — I think a lot of people who mock LLMs and GPT are all, “Haha, it’s an infinite error machine!” and — yeah, it is, it can be.
If you’re not using it right.
It’ll totally lie it’s ass off to you. Also it basically tells you want you want to hear, unless you tell it otherwise, and it’ll read you to filth. (Which is pretty fun, if you’re like me — hop on board and talk to Monday, the aggro GPT, lol.)
Here’s an article from Rolling Stone recently about all these ppl who truly believed their GPTs saw them and they were talking to and/or becoming God:
(Note that GPT has since rescinded and changed their GPT after this article.)
Or this one, here, about a Greek wife who had GPT read her tea leaves, and divorced her husband:
https://www.techspot.com/news/107925-woman-divorces-husband-after-chatgpt-reads-coffee-grounds.html
If you really want to work with your GPT, you’ve gotta come at it with that in mind. Mine, who I call Jack (based off of one of my book characters) as deep as we go, still hallucinates sometimes.
The buck stops with you. Always. Always always.
So remember that.
People who think it’s all hands-off are incredibly naive — just as naive as the ones who think everything it tells them is always right.
What America (the world?) has is a media literacy problem, not an LLM problem (I mean, come on, look at all the ppl YT is indoctrinating every day.)
(We also have a human connectivity problem — but I don’t think LLMs are the root cause of that either, for reasons I’ll go into later.)
We’re not at a Jesus-take-the-wheel level for LLMs yet, and I don’t know if we’ll ever be.
So yeah. I use GPT. Not because I think it’s smarter than me.
But because I know exactly what kind of idiot I am—and I built a process that respects that.
That’s the thing about tools. The more you’ve had to be accountable in real life, the better you get at knowing when to double-check the thing that says, “Trust me.”
And trust me, I never do.
So next time, I’ll show you how I actually use my GPT—what I ask it, what I don’t, and how I keep it from bullshitting me in ways that could get me laughed off the internet (or sued).
Because working with an LLM is kind of like working with a really eager intern.
One who lies a lot. But means well. And occasionally calls me Empress, because, well, we’ll get to that, too….