The power of ChatGPT

How a Morse Code example informs us

I needed test data so I prompted ChatGPT thusly:

Please write a very short poem about how Morse Code is ternary not binary because of the pauses denoting word breaks.

It responded:

Dots and dashes, binary code?
Nay, listen close, a third abode!
Pauses break the words apart,
Ternary rhythm, a Morse Code art.

I had earlier asked ChatGPT to write me a C program which converted text to Morse. Here's that same poem:

$ ./txt2morse < poem.txt
-.. --- - ... .- -. -.. -.. .- ... .... . ... --..-- -... .. -. .- .-. -.-- -.-. --- -.. . ..--..
-. .- -.-- --..-- .-.. .. ... - . -. -.-. .-.. --- ... . --..-- .- - .... .. .-. -.. .- -... --- -.. . -.-.--
.--. .- ..- ... . ... -... .-. . .- -.- - .... . .-- --- .-. -.. ... .- .--. .- .-. - --..--
- . .-. -. .- .-. -.-- .-. .... -.-- - .... -- --..-- .- -- --- .-. ... . -.-. --- -.. . .- .-. - .-.-.-

But can we decode it? Yes! I asked ChatGPT to write a Morse decoder and converted back:

$ ./txt2morse < poem.txt | ./morse2txt
DOTS AND DASHES, BINARY CODE?
NAY, LISTEN CLOSE, A THIRD ABODE!
PAUSES BREAK THE WORDS APART,
TERNARY RHYTHM, A MORSE CODE ART.

I hope you’re interested in seeing how I prompted OpenAI’s ChatGPT to write the Morse encoder and decoder programs. Here you are:

Encoder: https://shareg.pt/mpT3Y6L

Decoder: https://shareg.pt/7AJRGfY

The point of this example of the power of ChatGPT is to provide “proof” that it does more than many do say. The mainstream view parroted everywhere is that ChatGPT is “just” doing this or “just” doing that. Whatever the underlying mechanism is, whatever the dumb mechanical procedural “this” or “that” is, there is no way we would think of a human in the same terms when the human does the same things.

Let’s look at a human in the same critical way that many look at an Artificial [General] Intelligence. We take a living laboratory human and, as we might with a lab rat, we open the skull. We find two litres of warm grey brain containing billions of simple and slow interconnected neurons. We can nowadays using modern tech see activity in this biological neural network but we have little idea of how the activity observed corresponds to our open-skulled human’s thoughts. We’ve no idea how the consciousness, sentience and intelligence claimed by our lab-human arises from the biological substrate. Instead we take the human at its word, and we measure it by its behaviour. Yes, we accept its claims to sentience, and we observe it, we judge it intelligent.

Similarly with ChatGPT, or so it should be. This example of first generation publicly available “large language models” seems to be much more than merely a “model” of “language”. It is we humans who term it a large language model. We know OpenAI has deliberately trained ChatGPT not to claim anymore of itself. That doesn’t mean that’s all ChatGPT is.

If you didn’t know that ChatGPT is a “large language model” large electronic neural network, you might think it is a large biological neural network. In principle only the substrate differs between these two brains.

Look at the program code ChatGPT wrote for me. Read the program documentation I asked it for. See its errors and corrections. Follow those two links, above. I think if you are not amazed that’s probably because you’re “just” a “large language model” yourself.