i've joined @anomalyco to maintain sst and support @opencode
sst has always felt like the right abstraction - making complex but reliable infrastructure easy to use
it's an honor to work on that with people i've admired for years
first time working on an @EffectTS_ codebase and i already fell in love
i don't know if it's the guardrails or the spell it cast on the model, but i'm oneshotting stuff like never before
i don't know why this inspired me to write down all my thoughts about the topic, but here we go:
defined by economists, a job is a set of tasks. to complete a task, you need tools and the necessary skills
a common mistake is to compare AI to the industrial revolution. back then, advances in technology allowed the creation of tools to complete what were previously manual tasks. with the introduction of new tools, new skills were needed, both to design and operate them. that's why it created more jobs than it destroyed
but that’s not what’s going on with AI. researchers are not trying to automate a task; they’re trying to automate the skill behind all of them: intelligence
assuming they are able to do it, what’s left then?
many people point to taste or judgment, but i’m skeptical. if we define judgment as the ability to make the right decision, i don’t see why that wouldn’t just be a proxy for intelligence
another trait that’s impossible to automate is accountability. this is the one i thought would become the main bottleneck. for example, take the job of an accountant. are you paying them because they know the tax code, or because you want someone to blame when something goes wrong? knowing the tax code is essential, but so is their ability to convey trust
in other words: the AI can write the code, but is it going to merge the PR? this sounds reasonable until you start thinking about how smart these things could get. probably, at some point on the intelligence curve, we might start trusting AI more than a human (think whether you would trust a mediocre accountant or gpt-10)
i wrote all of this in a very assertive voice, but i'm as uncertain about it as anyone. some days it feels real, and others, pure science fiction
i don't know why this inspired me to write down all my thoughts about the topic, but here we go:
defined by economists, a job is a set of tasks. to complete a task, you need tools and the necessary skills
a common mistake is to compare AI to the industrial revolution. back then, advances in technology allowed the creation of tools to complete what were previously manual tasks. with the introduction of new tools, new skills were needed, both to design and operate them. that's why it created more jobs than it destroyed
but that’s not what’s going on with AI. researchers are not trying to automate a task; they’re trying to automate the skill behind all of them: intelligence
assuming they are able to do it, what’s left then?
many people point to taste or judgment, but i’m skeptical. if we define judgment as the ability to make the right decision, i don’t see why that wouldn’t just be a proxy for intelligence
another trait that’s impossible to automate is accountability. this is the one i thought would become the main bottleneck. for example, take the job of an accountant. are you paying them because they know the tax code, or because you want someone to blame when something goes wrong? knowing the tax code is essential, but so is their ability to convey trust
in other words: the AI can write the code, but is it going to merge the PR? this sounds reasonable until you start thinking about how smart these things could get. probably, at some point on the intelligence curve, we might start trusting AI more than a human (think whether you would trust a mediocre accountant or gpt-10)
i wrote all of this in a very assertive voice, but i'm as uncertain about it as anyone. some days it feels real, and others, pure science fiction