So I was bored and irritated by incessant buzzing of news about “GenerativeAI” being this, that, and the other, and I wrote a rant, and then decided to feed it to ChatGPT-4o, and this is what I got.
It's only paradoxical if you think that artificial neural networks can "have grasp" on things. It simply statistically determines which words need to appear in which order based on hundreds and thousands of works written by and selected by humans, in this case artificial intelligence researchers and practitioners (which indeed do have good grasp on these things).
Notice how in the response humans are referred as "we" because that's what was in the articles the ANN was trained on.
Yes, it simply agreed with the assertions being made.
Given contradictory input, it may well do the same.
Bottom line: Results can be coerced. There are no guarantees with regard to output. It displays remarkable language skill --- as long as the substance doesn't really matter.
The most astounding aspect is the number of supposedly intelligent, technical leaders who readily accept this as "intelligence".
Amusingly, the author write out Boyd's Observe -> Orient -> Decide -> Act loop almost exactly, although the author has it as Act -> Observe -> Orient -> Decide. Some of the 4 Boyd steps are broken into 2, so slightly more granular.
I wanted to highlight the 2 key components missing from "genAI", namely 1) continuous and real-time retraining and 2) high-level goal assessment mechanism. With the former not only missing but technologically unfeasible at this stage.
Without having these at least on the horizon it's ridiculous to be talking about "emerging intelligence" and "AGI being just around the corner".
6 comments
[ 3.1 ms ] story [ 28.4 ms ] threadParadoxically, this sounds like LLMs have a better grasp of LLMs than many humans do.
Notice how in the response humans are referred as "we" because that's what was in the articles the ANN was trained on.
Given contradictory input, it may well do the same.
Bottom line: Results can be coerced. There are no guarantees with regard to output. It displays remarkable language skill --- as long as the substance doesn't really matter.
The most astounding aspect is the number of supposedly intelligent, technical leaders who readily accept this as "intelligence".
Without having these at least on the horizon it's ridiculous to be talking about "emerging intelligence" and "AGI being just around the corner".
Edited for clarity