Why does this blogpost read as it was written a year ago? He speaks at getting 100% on a test if we let him google the answers, while LLMs are now beating students at Math Olympiad and even most adult specialists at various other intellectual competitions. Time to update knowledge cutoff point of your wetware you meatbag.
These are not the reasons why I hate AI. I hate AI because we still don't have universal housing or healthcare and AI's direct endgame is to eliminate every job that isn't CEO. And what a fucking drab world that will be for all the displaced workers.
> Whenever I’m critical of anything GenAI, without fail I get asked the same question. “do you think every major CEO could be wrong?”
I think one of the most helpful mental models about unanimity is, are the decisions independent, or unified? In theory if you do a literature review, you’re summarizing different research experiences, but if you’re 100 CEO’s, it’s all the same data they’re looking at, it’s really only one opinion, and prediction is hard, especially about the future.
If nothing else I think every major CEO lives in a different would than I do and has very different motivations than I do. And most of those motivations are in direct opposition to a healthy society. So yes, I do think every major CEO is wrong. If for no other reason but because of short term, earnings report driven thinking.
14 comments
[ 3.6 ms ] story [ 32.0 ms ] threadI think one of the most helpful mental models about unanimity is, are the decisions independent, or unified? In theory if you do a literature review, you’re summarizing different research experiences, but if you’re 100 CEO’s, it’s all the same data they’re looking at, it’s really only one opinion, and prediction is hard, especially about the future.
If nothing else I think every major CEO lives in a different would than I do and has very different motivations than I do. And most of those motivations are in direct opposition to a healthy society. So yes, I do think every major CEO is wrong. If for no other reason but because of short term, earnings report driven thinking.
There you go. Saved you from having to read 2000 words of generic ranting.
In other words, the output from current LLMs will always contain a significant amount of arbitrary, random data --- aka BS.
This; combined with their high operating costs, severely limits the applicability of LLMs.
Traditional computing offers highly accurate results at low cost. LLMs turn this upside down and offer questionable results at high cost.