so, certainly possible to learn a bit from this for a newbie,
and pretty darn good for ChatGPT version 1.0
It is not a good idea to try to learn scripting from ChatGPT.
It is unable to calculate. It can output calculations that were present in its training data. If the calculation did not exist, it cannot calculate it either. It fails at multiplications of multi-digit numbers, even at the addition of larger numbers.
It cannot solve logical problems if they were not in the data sets. It fails at logical problems that 10 year old kids can solve.
Sorry,
no mind, no interligence at all. This is not surprising when you look at
how language models work.
If you roughly understand
language models, you can also see why
facts are wrong if there is no data basis. The method is based on statistical predictions of what the following words most likely are. That is, if the correct answer was not found in the training data, you will get a wrong (
but correct-looking) answer. Not exactly ideal for a beginner who can't see that.
There have already been some attempts to have ChatGPT programs written. Many are simply
buggy, just strewn together according to the language model. This applies in particular to a scripting language like LScript, which is certainly not one of the most common due to its special use.
This makes it very difficult, if not impossible to create really
new programs that work.
It is (obviously) similar to the
diffusion models for imaging. If you roughly understand the process, it immediately becomes clear that the creation of these images has nothing to do with the process of living beings called
creativity.
We still don't have a comprehensive understanding of how the brain works. Therefore, it is difficult to predict whether today's approaches to AI will ever lead to something similar or not. What is clear, however, is that current programs lack any kind of intelligence, understanding or creativity. I don't mean the intelligence and creativity that some programmers put into creating it.
I think it will take some time before the hype cools down and people understand what's really going on here. Then you will be able to use the strengths of this models sensibly (on the less sensational side, there are already applications that use these models practically).
However, it's nice to try stuff like this and show it off.
ciao
Thomas