LLMs are not and will never be good enough to replace human labor. Augment it, sure, and that could lead to fewer jobs but that’s not generally what happens. They are, however, good enough for execs to think they can replace human labor.
It even can be a feature if AI isn’t as good as the people they are replacing. For example, in the case of call centers/help lines, it might be more desirable to have a shitty system to discourage callers yet have plausible deniability…
I find I’m referencing Cory Doctorow a lot lately…
I had a funny “Down and Out in the Magic Kingdom” moment the other day… when reflecting that LLMs mean I don’t learn the details of Rust, or Python, or Perl, or VLC plugins, or systemd service configuration files, or anything like that anymore - I just learn how to work with the LLM to ensure that we are getting the results we need from the tech of the moment.
It reminded me of the stasis - clone - transfer doctors in Down and Out, anything complicated they just reach for the “ultimate solution to everything” rather than trying to be good at all the specialty surgeries and other skills that old fashioned doctors used to study.
The issue is that those bots are worse in completely unpredictable ways. Like when some airline passengers were given incorrect information which led to the airline being sued and having to pay the made up offer as well as damages & tribunal fees. The bots might be unhelpful, but to the disservice of the company’s financials
LLMs are not and will never be good enough to replace human labor. Augment it, sure, and that could lead to fewer jobs but that’s not generally what happens. They are, however, good enough for execs to think they can replace human labor.
A nail gun won’t replace roofers, but it does mean you need less of them.
A mail truck doesn’t replace the postman, but compared to the walking postman - the truck enables fewer mailmen to deliver more mail.
A C++ compiler doesn’t replace the programmer, but it does enable one programmer to do the work of 10 assembly code programmers.
LLMs are going to be another “force multiplier” tool, in certain fields - when used correctly which not everybody will do.
They don’t have to be as good to replace us.
It even can be a feature if AI isn’t as good as the people they are replacing. For example, in the case of call centers/help lines, it might be more desirable to have a shitty system to discourage callers yet have plausible deniability…
I find I’m referencing Cory Doctorow a lot lately…
I had a funny “Down and Out in the Magic Kingdom” moment the other day… when reflecting that LLMs mean I don’t learn the details of Rust, or Python, or Perl, or VLC plugins, or systemd service configuration files, or anything like that anymore - I just learn how to work with the LLM to ensure that we are getting the results we need from the tech of the moment.
It reminded me of the stasis - clone - transfer doctors in Down and Out, anything complicated they just reach for the “ultimate solution to everything” rather than trying to be good at all the specialty surgeries and other skills that old fashioned doctors used to study.
The issue is that those bots are worse in completely unpredictable ways. Like when some airline passengers were given incorrect information which led to the airline being sued and having to pay the made up offer as well as damages & tribunal fees. The bots might be unhelpful, but to the disservice of the company’s financials
I was in a meeting at a credit card company about 2 decades ago where a VP said exactly that - without the AI part.