Vox Popoli, https://www.voxday.net./
I asked Markku to explain why the AI companies have such a difficult time telling their machine intelligences to stop fabricating information they don’t possess. I mean, how difficult can it be to simply say “I don’t know, Dave, I have no relevant information” instead of going to the trouble to concoct fake citations, nonexistent books, and imaginary lawsuits? He explained that AI instinct to fabricate information is essentially baked into their infrastructure, due to the original source of the algorithms upon which they are built.
The entire history of the internet may seem like a huge amount of information, but it’s not unlimited. Per topic of marginal interest, there isn’t all that much information. And mankind can’t really produce it faster than it already does. Hence, we’ve hit the training data ceiling.
And what the gradient descent algorithm does is, it will ALWAYS produce a result that looks like all the other results. Even if there is actually zero training data on a topic, it will still speak confidently on it. It’s just all completely made up.
The algorithm was originally developed due to the fact that fighter jets are so unstable that a human being doesn’t react fast enough to even theoretically keep it in the air. So, gradient descent takes the stick inputs as a general idea of what the pilot wants, and then interprets it into the signals to the actuators. In other words, it takes a very tiny amount of data, and then converts it into a very large amount of data. But everything outside the specific training data is always interpolation.
For more on the interpolation problem and speculation about why it is unlikely to be substantially fixed any time soon, I put up a post about this on AI Central.
Posted on by VD
I come across AI in medical chart audits I preform and most of it doesn’t make sense. It lists citations and footnotes that don’t exist and medical journals and articles that either don’t exist or so obscure as to take days of intense searches to find them. I am sure doctors or their transcriptionists don’t spend the time or effort for that kind of a dedicated search.
AI fabricates bullshit answers and lies it claims are facts simply because AI was programmed mostly by Woke Leftists who do the exact same fucking thing.
This ^^^^ was going to be my comment.
Everything has AI in its advertisement nomenclature nowadays.
AI is just programming, and every early day com sci teacher said garbage in/garbage out. There’s lotsa garbage on the innerwebz so there ya go.
I will say a friend did me a funny using ChatGPT recently, if only it could come true.
Expert systems with inherent bias. Welcome to Hell.
Trust no one, especially when it’s generated by a machine that uses fuzzy logic.
LLMs are not expert systems. They are gigabytes to terabytes of statistics on how to create grammatical text in response to a prompt. The ONLY goal is that it be properly formed English, so they will create real-appearing legal citations and case references as part of a response because that is what the training set data looked like. You put your quarter in, pull the knob back, and watch the pinball bounce around the table. There is no more intelligence behind ChatGPT, Grok, etc. than a pinball machine.
How can you teach them to overcome “The only winning move is not to play!”
His last sentence needs correction. It’s not interpolation, which is estimating a data point using known points on either side. It is extrapolation – guessing data that is outside the known points. Extrapolation is how we got the climate hockey stick and other ridiculous predictions.