METR study is brutal. 20% faster vs actually 19% slower - that's not rounding error territory.
Productivity measurement is a mess though. Spent three weeks last spring chasing metrics on a "microservices will save us" project. Spoiler: they didn't. Could've been TDD, could've been GraphQL, could've been AI - same story. We suck at measuring ourselves.
But that asymmetry thing - yeah, that's real. Spammer generates 10K bot accounts, never checks if they work right. I write code with Copilot, gotta verify every function or I'm debugging hallucinated bugs at 3am. Harm scales unlimited. Benefit hits human bandwidth ceiling. Hard ceiling.
Slot machine comparison works for coding, breaks everywhere else. I use it for boring CRUD boilerplate, ticket categorization, email templates I'd write anyway. Not life-changing. Not gambling addiction either. Just... mildly useful sometimes? Context dependent.
Hallucinations aren't fixable - author's right there. But calculators can give you garbage output too if you fat-finger 7 instead of 4. Still use them. Just don't blindly trust them.
That is a certainty. I was once asked to calculate how much time we would save through our companies code reuse program. I read all the material on estimating savings, but then proved it was all ridiculous.
I came across a study that attempted to estimate how long it took to build libraries that had already been built. In this case, there were no unknowns, you had the entire code. Estimates were off by orders of magnitude. If we can't estimate the work when the work is already done, how could we ever estimate the work when we know less?
> What if the hallucinations that matter aren’t those from the AI, but the humans who are hallucinating that AI is intelligent, conscious, creative, or even competent at what it does?
Yes, to me there's a constant theme of LLM "AI" discussions, where a bunch of humans are trusting too much in a subconscious bias which has, after many good years, become problematic. A kind of pareidolia of conversations and minds, rather than pictures and faces.
The same process that makes people think "Bot talking to me, Bot is a real mind in silicon" can be trivially twisted to fit a "conversation" with Count Dracula, blood-thirsting lord of the night, locked in a coffin with a laptop. The difference is people have more mental-armor against thinking Dracula could be real.
METR study is brutal. 20% faster vs actually 19% slower - that's not rounding error territory.
Productivity measurement is a mess though. Spent three weeks last spring chasing metrics on a "microservices will save us" project. Spoiler: they didn't. Could've been TDD, could've been GraphQL, could've been AI - same story. We suck at measuring ourselves.
But that asymmetry thing - yeah, that's real. Spammer generates 10K bot accounts, never checks if they work right. I write code with Copilot, gotta verify every function or I'm debugging hallucinated bugs at 3am. Harm scales unlimited. Benefit hits human bandwidth ceiling. Hard ceiling.
Slot machine comparison works for coding, breaks everywhere else. I use it for boring CRUD boilerplate, ticket categorization, email templates I'd write anyway. Not life-changing. Not gambling addiction either. Just... mildly useful sometimes? Context dependent.
Hallucinations aren't fixable - author's right there. But calculators can give you garbage output too if you fat-finger 7 instead of 4. Still use them. Just don't blindly trust them.
> We suck at measuring ourselves.
That is a certainty. I was once asked to calculate how much time we would save through our companies code reuse program. I read all the material on estimating savings, but then proved it was all ridiculous.
I came across a study that attempted to estimate how long it took to build libraries that had already been built. In this case, there were no unknowns, you had the entire code. Estimates were off by orders of magnitude. If we can't estimate the work when the work is already done, how could we ever estimate the work when we know less?
> What if the hallucinations that matter aren’t those from the AI, but the humans who are hallucinating that AI is intelligent, conscious, creative, or even competent at what it does?
Yes, to me there's a constant theme of LLM "AI" discussions, where a bunch of humans are trusting too much in a subconscious bias which has, after many good years, become problematic. A kind of pareidolia of conversations and minds, rather than pictures and faces.
The same process that makes people think "Bot talking to me, Bot is a real mind in silicon" can be trivially twisted to fit a "conversation" with Count Dracula, blood-thirsting lord of the night, locked in a coffin with a laptop. The difference is people have more mental-armor against thinking Dracula could be real.