The Damned Dangers of Ultracrepidarianism and AI
One of the main problems with AI - and specifically LLMs - is that to understand what’s going on, you need to have an understanding of a large number of fields: human cognition, stochastic modeling, machine learning, stock market bubble behavior, circular lending, hardware lifetimes, etc.
Almost every expert has deep knowledge in one of those fields - some even have two. But those of us wandering tinkerers, who have a small practical knowledge of a number of them, know enough to call bullshit when an expert rambles outside their domain (ultracrepidarianism) and says something demonstrably false, but not enough to be confident of the whole picture.
Hell, so much of the way industry leaders talk about AI wanders into territory that is either deliberately fabulist and self-serving, clearly and inescapably technically wrong, or disturbingly sociopathic that they’re useless. Especially since most of their “good outcomes” are actually dystopian. And that’s not just execs, but tech leaders and watch dogs.
There’s just so much I don’t know. I don’t know what ‘inference” really means, other than clustering similarities, I don’t know how any of the “breakthroughs” on cost are actually being tested by the Chinese. I don’t know what the socioeconomics of ‘winning’ on the AI front by our government subsidizing it and then ‘losing’ on energy infrastructure and storage means. I don’t know whether the patterns of thinking from ‘Thinking Fast and Slow’ applies to AI and if so, whether that requires an ontological backbone and integration into pattern analysis… and on and on.
But like Rilke said, I try to love the questions themselves, like locked rooms or books written in a foreign tongue.
So in that spirit - what are all the burning questions you wish you had answers to?
And just to make this perfectly clear in advance, this is not a f*cking AI post.