LLMs produce proper English and offer reasonable follow-ups. Funnily enough, recent models are less good at this, but a good harness and skillset can do wonders.
This is in contrast with my colleagues from xxxxx [0], who appear to be basically immune to learning to read, write, speak, or understand English any reasonably, despite working at a multinational company explicitly requiring so. They also keep insisting on pointless meetings where I get to tell them the exact same things I already told them in writing, except now it is across both of our accents, and through their terrible microphone and internet connection. Mind you, I do listen myself back sometimes, so this is not me being gleefully self-unaware either.
Nothing against them personally, but I truly despise working with these people. They have zero edge over even something like Luna, and will 100% be replaced wholesale once management stops pretending, unless they start using LLMs hard. Something they're unable to do, for the same reasons they're terrible to work with. [1]
And there's no amount of doorman fallacy invocations that will save this. They're cheap, sure, but again, even the cheap models meet or beat them. They've been meat proxies even before LLMs, and the amount of instructions they need to be given will make even the dumbest LLM fly through things. Even though they're hired in swarms, not unlike how people try to save terrible LLMs, that only takes you so far. They don't grow, or grow at glacial speeds. They're cooked.
If you can offload a task to Sonnet, you met them in price, and beat them dozens of times over in quality. Very inhumane way to put it, but it's the reality already. And the bar is rising.
[0] Censored at will because this is not a race thing, but bottom of the barrel HR practices and market realities.
[1] In true terrible management fashion though, it is entirely possible that they'll simply fire the more capable and more expensive people instead, falsely expecting that these other folks will wonderfully use LLMs to just figure things out. Now that will be quite the fireworks show!
A lot of engineers are gonna hate this, but the domain that LLMs will have the hardest time dominating is people skills
And when they do dominate that we’re going extinct within 10 years so it’s a good bet to work on those.
Romance chatbots say what?
LLMs produce proper English and offer reasonable follow-ups. Funnily enough, recent models are less good at this, but a good harness and skillset can do wonders.
This is in contrast with my colleagues from xxxxx [0], who appear to be basically immune to learning to read, write, speak, or understand English any reasonably, despite working at a multinational company explicitly requiring so. They also keep insisting on pointless meetings where I get to tell them the exact same things I already told them in writing, except now it is across both of our accents, and through their terrible microphone and internet connection. Mind you, I do listen myself back sometimes, so this is not me being gleefully self-unaware either.
Nothing against them personally, but I truly despise working with these people. They have zero edge over even something like Luna, and will 100% be replaced wholesale once management stops pretending, unless they start using LLMs hard. Something they're unable to do, for the same reasons they're terrible to work with. [1]
And there's no amount of doorman fallacy invocations that will save this. They're cheap, sure, but again, even the cheap models meet or beat them. They've been meat proxies even before LLMs, and the amount of instructions they need to be given will make even the dumbest LLM fly through things. Even though they're hired in swarms, not unlike how people try to save terrible LLMs, that only takes you so far. They don't grow, or grow at glacial speeds. They're cooked.
If you can offload a task to Sonnet, you met them in price, and beat them dozens of times over in quality. Very inhumane way to put it, but it's the reality already. And the bar is rising.
[0] Censored at will because this is not a race thing, but bottom of the barrel HR practices and market realities.
[1] In true terrible management fashion though, it is entirely possible that they'll simply fire the more capable and more expensive people instead, falsely expecting that these other folks will wonderfully use LLMs to just figure things out. Now that will be quite the fireworks show!
> You have to beat the models at something
And for the rest of your career* you'll be running on a treadmill trying to stay ahead of them!
(* probably like two years if you're one of the lucky ones)
Finally someone is asking the right questions.
and then teach ai how to beat you
[dead]