 |
| From the IEEE article |
I have not posted for a long time for several reasons... Busy with life, etc.
This post from IEEE: "AI Mistakes Are Very Different Than Human Mistakes We need new security systems designed to deal with their weirdness" is simply beyond belief.
There have been numerous research articles, like "Apple’s latest study proves that AI can’t even solve basic grade-school math problems" (paper here), showing that if you train you AI with a bunch of kid stories, .e.g., "Little Johnny has $1.00 and Suzy is selling eggs for $.25..." you can ask questions about little Johnny and little Suzy but you can't ask the AI to infer what happens if the names change.
So the AI is just essentially doing a JPEG-style compression on the billions of parameters and making them into a google-like search environment. No reasoning at all.
Consider the IEEE article in this context.
Of course the AI will make mistakes - it can't distinguish between Johnny and Frank - because it can't think and only mimics the input data weakly.
Either fix the AI or don't use it, but don't add another stupid layer of useless software on top of it to "fix it."
If you can just "fix it" why bother using the AI at all...?
