"If you want a picture of the future, imagine a bunch of LLMs making decisions on all of our behalves, forever."
Will they be making actual decisions for us? Or will they be having endless pointless conversations with each other on our behalves, emails full of bland vocabulary arranged into ambiguous meanings?
Software is nothing if not a mechanical abstraction for bureaucracy.
How many times have you heard, "I'm sorry, The System won't let me do that".
Imagine now, that all of these The Systems are singing the siren song of the zeitgeist in perfect unison. I can guarantee you that any company that's performing customer support at scale is absolutely scrambling to replace all of their offshore support reps with an LLM. Amazon in particular, who likely does more B2C volume than any other entity in the world, and who in particular has already released tooling to their Web Services product to allow anyone to build on their own LLM infrastructure, is almost certainly guaranteed to do this.
Everyone else is going to follow suit as quickly as possible. There's too much of an obvious financial upside to doing this, for it to not happen.
Here's what it boils down to: the human race has been actively surrendering its agency as quickly as possible since the Greeks decided that they didn't want a true democracy anymore, and we all know how that ended. The incentive here is simple - when you surrender your agency, you also surrender your responsibility for the outcome of your actions, because they aren't your actions anymore.
Before LLMs, all we had were politicians and the DSM-V; now that we have created machines that can give the impression of having compiled all of the world's knowledge (fools that we are, as if "all of the world's knowledge" could ever be contained in "whatever's been on the Internet since 1995"), queryable through natural language - how long do you think it's going to be before our governments start surrendering THEIR agency?
You can take this in any dystopian direction that you want; there are too many to enumerate, and quite frankly I have no interest in scaremongering so I'm not going to provide examples.
See exactly what you mean. Great example. But context is everything. If using your example “ ___ of the population commit ___ of the crime ” and LLM tells me that I am ____ I laugh. If a LLM tells me “we will bury you” I laugh. If the example above has the force of law then it is something else. If Nikita Khruchev tells me “We will bury you” and you are in the middle of a cold war it means something else and even the people who thought he was talking about nuclear war were wrong. But you are correct to be worried about this and the loss of agency.
Right now it is hard to see where it will go. Getting rid of customer service jobs I can see. How invasive will it be or how much damage and what unintended the side effects are anyone’s guess. You are not fear mongering by bringing up alternative paths this could take. Like any thinker you are evaluating all the possible outcomes.
Not having Alexa and disabling Siri (yeah I know big brother is watching AND listening), not being trendy, not ordering online I am hoping this takes a while to reach me. It will in annoying ways at first. Stores will try to use it instead of a person to answer calls or take orders at my pizza place. Where it expands we shall see.
I see what you mean. I don't work in software and I haven't tried using ChatGPT so the only exposure I get to this subject is through Jack and now this article. I'm able to write my own emails and I've found greater happiness and personal satisfaction when I take responsibility for my actions (including my stupid mistakes) so I don't see any personal benefit to using an LLM. I have tried to get AI to make some art I could use as covers to my stories, but what Night Cafe produces is deeply unsatisfying (and often profoundly disturbing), but that's a little different from an LLM (how different I'm not sure, since I don't know much about the programming behind art AI and writing AI). Thank you for your response, I do appreciate it. I was not trying to be glib in my own comment, I just imagined a million million AIs sending each other a billion billion meaningless messages based on my own limited experience.
"replace all of their offshore support reps with an LLM"
I made precisely this prediction over at Locklin's blerg. Seems like low hanging fruit since they run off a script for the most part in low level tech support.
Elsewhere I said places handling tickets like ServiceNow would have first stop techs replaced with an LLM offering.
What I find particularly troubling is the gold-rush to replace _all kinds of_ entry-level positions with an LLM.
"Let's replace all of the entry-level positions with LLMs and let the senior engineers captain the ship" sounds great, until you realize that senior engineers were once juniors - in a field that already has an astoundingly-high barrier to entry as it is.
If we pull up the ladder, there won't _be_ any battle-hardened master craftsmen once the current generation leaves the workforce.
Part of me feels the need to corkscrew down into this; but the more I look, the worse it gets, in every single direction imaginable.
The accounting profession is currently doing this at the big 4. The big 4 trickles down to the rest of the economy. No one is going into the accounting field because the entry level pay is crap and we can't find experienced accountants because of it. But hey, at least the partners at the big 4 are getting fat off the seed corn.
That's a good insight about human capital (a term I dislike but w/e).
I've heard lots of excitement that boils down to people losing their jobs but it's always marketed as saving time or enabling people. We might just enable ourselves into an ever further stratified society.
I am currently our low level programmer's go to guy for most questions and it'll be interesting if I can hand off my debugging talent which is one of the few things I consider myself good at.
I need to cut back on bombast, but one could think of AI as metastatic antihuman. It can't make a prokofiev symphony, but it could be used to generate endless streams of ariana grande-tier top 40--with even less coherent lyrics. It can't write the software equivalent of Signal or WhatsApp, but could write an insecure equivalent easily pwned by a hobbyist hacker. It can write what you might refer to as poetry, but so far I've never seen it express the poetry of soul, the tragedy of loss, the comic in the profane.
And we're introducing this metastatic-NPC-meme software into a society with a gigantic mass of both unemployed and essentially unemployable people - people even *less* capable than AI of producing valuable art or work product.
"In the beginning was the Word, and the Word was with God, and the Word was God."
Pick your God carefully.
> Clippy breathing heavily in the background
Cleverness is its own end, and having failed to find a path to true AI, man is endeavoring to create god in his own image with LLM.
"If you want a picture of the future, imagine a bunch of LLMs making decisions on all of our behalves, forever."
Will they be making actual decisions for us? Or will they be having endless pointless conversations with each other on our behalves, emails full of bland vocabulary arranged into ambiguous meanings?
Software is nothing if not a mechanical abstraction for bureaucracy.
How many times have you heard, "I'm sorry, The System won't let me do that".
Imagine now, that all of these The Systems are singing the siren song of the zeitgeist in perfect unison. I can guarantee you that any company that's performing customer support at scale is absolutely scrambling to replace all of their offshore support reps with an LLM. Amazon in particular, who likely does more B2C volume than any other entity in the world, and who in particular has already released tooling to their Web Services product to allow anyone to build on their own LLM infrastructure, is almost certainly guaranteed to do this.
Everyone else is going to follow suit as quickly as possible. There's too much of an obvious financial upside to doing this, for it to not happen.
Here's what it boils down to: the human race has been actively surrendering its agency as quickly as possible since the Greeks decided that they didn't want a true democracy anymore, and we all know how that ended. The incentive here is simple - when you surrender your agency, you also surrender your responsibility for the outcome of your actions, because they aren't your actions anymore.
Before LLMs, all we had were politicians and the DSM-V; now that we have created machines that can give the impression of having compiled all of the world's knowledge (fools that we are, as if "all of the world's knowledge" could ever be contained in "whatever's been on the Internet since 1995"), queryable through natural language - how long do you think it's going to be before our governments start surrendering THEIR agency?
You can take this in any dystopian direction that you want; there are too many to enumerate, and quite frankly I have no interest in scaremongering so I'm not going to provide examples.
Just consider that at some point,
___ of the population commit ___ of the crime.
may be used to decide your fate.
See exactly what you mean. Great example. But context is everything. If using your example “ ___ of the population commit ___ of the crime ” and LLM tells me that I am ____ I laugh. If a LLM tells me “we will bury you” I laugh. If the example above has the force of law then it is something else. If Nikita Khruchev tells me “We will bury you” and you are in the middle of a cold war it means something else and even the people who thought he was talking about nuclear war were wrong. But you are correct to be worried about this and the loss of agency.
Right now it is hard to see where it will go. Getting rid of customer service jobs I can see. How invasive will it be or how much damage and what unintended the side effects are anyone’s guess. You are not fear mongering by bringing up alternative paths this could take. Like any thinker you are evaluating all the possible outcomes.
Not having Alexa and disabling Siri (yeah I know big brother is watching AND listening), not being trendy, not ordering online I am hoping this takes a while to reach me. It will in annoying ways at first. Stores will try to use it instead of a person to answer calls or take orders at my pizza place. Where it expands we shall see.
I see what you mean. I don't work in software and I haven't tried using ChatGPT so the only exposure I get to this subject is through Jack and now this article. I'm able to write my own emails and I've found greater happiness and personal satisfaction when I take responsibility for my actions (including my stupid mistakes) so I don't see any personal benefit to using an LLM. I have tried to get AI to make some art I could use as covers to my stories, but what Night Cafe produces is deeply unsatisfying (and often profoundly disturbing), but that's a little different from an LLM (how different I'm not sure, since I don't know much about the programming behind art AI and writing AI). Thank you for your response, I do appreciate it. I was not trying to be glib in my own comment, I just imagined a million million AIs sending each other a billion billion meaningless messages based on my own limited experience.
"replace all of their offshore support reps with an LLM"
I made precisely this prediction over at Locklin's blerg. Seems like low hanging fruit since they run off a script for the most part in low level tech support.
Elsewhere I said places handling tickets like ServiceNow would have first stop techs replaced with an LLM offering.
What I find particularly troubling is the gold-rush to replace _all kinds of_ entry-level positions with an LLM.
"Let's replace all of the entry-level positions with LLMs and let the senior engineers captain the ship" sounds great, until you realize that senior engineers were once juniors - in a field that already has an astoundingly-high barrier to entry as it is.
If we pull up the ladder, there won't _be_ any battle-hardened master craftsmen once the current generation leaves the workforce.
Part of me feels the need to corkscrew down into this; but the more I look, the worse it gets, in every single direction imaginable.
The accounting profession is currently doing this at the big 4. The big 4 trickles down to the rest of the economy. No one is going into the accounting field because the entry level pay is crap and we can't find experienced accountants because of it. But hey, at least the partners at the big 4 are getting fat off the seed corn.
That's a good insight about human capital (a term I dislike but w/e).
I've heard lots of excitement that boils down to people losing their jobs but it's always marketed as saving time or enabling people. We might just enable ourselves into an ever further stratified society.
I am currently our low level programmer's go to guy for most questions and it'll be interesting if I can hand off my debugging talent which is one of the few things I consider myself good at.
When it comes to programming I don't think llm can explain why something is wrong given how it generates textd
At least the English might be better!
Dammit, it’s “that,” not “dart!” 🙄
Magnificent article. Thank you for thinking through and executing such an important message
Excellent piece. More, please!
I need to cut back on bombast, but one could think of AI as metastatic antihuman. It can't make a prokofiev symphony, but it could be used to generate endless streams of ariana grande-tier top 40--with even less coherent lyrics. It can't write the software equivalent of Signal or WhatsApp, but could write an insecure equivalent easily pwned by a hobbyist hacker. It can write what you might refer to as poetry, but so far I've never seen it express the poetry of soul, the tragedy of loss, the comic in the profane.
And we're introducing this metastatic-NPC-meme software into a society with a gigantic mass of both unemployed and essentially unemployable people - people even *less* capable than AI of producing valuable art or work product.