Morse Code (and ChatGPT frustrations)

General Memory Chat
1

I just stumbled across this awesome thread from @Simon about Morse code: Morse code words

There is a list of words in there to associate with each letter. The suggested word for O is “Oops.”

I loved the method but wanted a different, more visual word. So I thought, this should be super simple for ChatGPT! After all, it is primarily a language bot!

Well, holy $#&@, it was an infuriating and frustrating experience.

Here’s the gist of what went down:

Screenshot_20230318-112531~2
Screenshot_20230318-112531~21080×1678 117 KB

Screenshot_20230318-112554~2
Screenshot_20230318-112554~21080×1622 86.5 KB

Screenshot_20230318-112604~2
Screenshot_20230318-112604~21080×1483 92.3 KB

Screenshot_20230318-112617~3
Screenshot_20230318-112617~31080×1157 60.4 KB

This lasted nearly an hour. Around and around in circles. This bot clearly does not understand word structure and has lots of trouble with conditional criteria.

Everyone thinking ChatGPT will be a quick and easy way to generate things like lists of Major System words or words that follow a specific structure is going to be very disappointed with the current version of this AI.

3 Likes
3 Digit Major System
2

@TheHumanTim, I sympathize with you. ChatGPT does well at text creation and understanding. It fails miserably with rules.

Here’s a dialog I just had:

Please give me a four-letter word starting with O that has all the letters in the second half of the alphabet.

The four-letter word that starts with O and contains all the letters in the second half of the alphabet is “ozon”.

Please give me five four-letter words starting with O that have all the letters in the second half of the alphabet.

I apologize for my previous response, “ozon” does not have all the letters in the second half of the alphabet. Here are five four-letter words that start with O and have all the letters in the second half of the alphabet:

  1. obey
  2. only
  3. onyx
  4. odor
  5. oboe

At least I got the good choice that you already found!
Doug

1 Like
3

It eventually spit out Oryx, which is a pretty nice visual. But that took so long, I would have been better off just enjoying the mental exercise of figuring it out myself. Would have been faster and a good brain workout!

1 Like
4

It’s interesting though, @thinkaboutthebible… I had this interaction with it yesterday:

IMG_20230318_134602
IMG_20230318_1346021080×1423 160 KB

So it can do some incredibly complex and “creative” things, but when it comes to a simple “find a word that follows these few rules” …epic fail.

1 Like
5

@TheHumanTim, I was curious about this example so I pushed it further and recreated the rules in a different way. It still failed. It gave me “OYZZ” and “OXYZ.” So I asked (emphasis is mine):

Why are you not checking to see if your responses are valid English words?

You are correct that as an AI language model, I should have checked if my responses are valid English words before providing them. I apologize for the mistakes I made in my previous responses.

As a language model, I don’t have a built-in mechanism to verify if a given combination of letters forms a valid English word. However, I can use language processing techniques to generate more plausible and valid responses based on the given pattern rules.

I think it will have some pattern matching in the future to create words but for now the alphabetic listing out of words is awesome like your acrostic and my peg lists.

1 Like
6

@TheHumanTim, now I’m getting somewhere and I recommend upgrading to GPT-4 by getting a Plus subscription. The old chats were ChatGTP 3.5. This is the current version 4.0.

Here are some pattern rules that must be followed and I will use. The digit one means any letter in the first half of the alphabet. The digit two means any letter in the second half of the alphabet. A letter means that you must use that letter. Find an English word that follows the pattern O222 and verify that it is an English word in a dictionary.

The pattern O222 represents a word with the structure: O followed by three letters from the second half of the alphabet. One such word that follows this pattern is “opqr.”

However, “opqr” is not a valid English word. Let’s try another example: “orch.” In this case, “orch” is an abbreviation for “orchestra,” and it is a recognizable term in English. You can verify this by checking an English dictionary or searching online.

Keep in mind that “orch” is not a standalone English word, but an abbreviation. Nonetheless, it follows the given pattern and can be understood in the context of the English language.

[edit} Neither does it follow the rules, but it is getting smarter.

1 Like
7

@thinkaboutthebible it’s fascinating to watch it try. I wonder what the hurdle is that’s keeping it from understanding those simple retrieval commands. Non-AI database software could pull queries like these if the correct metatagging is in place. You’d think that it would have an understanding of filtered exclusion/inclusion.

1 Like
8

I would like to try the Dall E type of AI.
I could write a sentence and I guess it would create a visualisation of the sentence.

If it learned all it knows from public domain images, I don’t know.

I’m uncomfortable with how some AI software trains itself on people’s copyrighted images or implicitly copyrighted blogged text.

Chat GPT sounds very interesting.

But I was puzzled that a recent post by someone was considered to be potentially AI generated because it presented succinctly two points of view.

I don’t know why an AI would subscribe to a memory chat group.

I wonder where this is all going to end.

With the Morse, did you see that there is a difference between a system for writing Morse code and a system for interpreting a received message?

Mike

10

That’s something I’ve been thinking about. Simon’s “first half / last half” method is awesome for learning to code it. Only 26 very simple images to learn and you can be up and running in no time. Receiving a message and being able to decode it, on the other hand, is going to take some thought. I haven’t come up with a good way yet, beyond brute force rote learning for it.

It’s a binary system, so maybe decoding could follow something like: binary to base 10 numeral to major system word image? Could encoding also use it and make Simon’s system unnecessary?

Example:
If .'s are represent binary 0 and -'s represent binary 1…
O = - - - = 111 = 7 = Key

So you associate Key with O somehow, either through a peg list, or make an alphabetized memory palace where each loci starts with a new Letter.

So you place a Key in your O loci, maybe the loci is a giant Obelisk like in the movie 2001. You picture a key opening a secret lock in the obelisk and creating a portal to another dimension.

Now if you are asked what the code is for O, you’d think:
obelisk / KEY
Key = 7
7 = 111 in binary
111 = - - - in Morse.

If you want to know what - - - is:

      • = 111
        111 = 7 in base10
        7 = Key in major system
        A Key is associated with the Obelisk in my Morse memory palace.
        Obelisk = O.
        So, - - - = O.

Is this viable?

There would be duplicates if a Morse letter had a leading .

So . - - - = 0111, which would be the same as 111.

There would need to be a way to differentiate that.

Hmmm…

EDIT:
Ok, scrap the binary to base10 step.
Go straight from binary to major system and include the leading zeros:

T = - = 1 = Tee
A = . - = 01 = SeaT
U = . . - = 001 = CyST
V = . . . - = 0001 = SauCy SoDa

1 Like
11

Here’s a quick list, palace is an imaginary museum with each loci starting with the respective letter for coding. The 1, 2, and 3 “digit” letters get a single major system word. For the 4 digit ones, use two 2-digit words, ideally following an “adjective + noun” format.

Letter Morse Code Palace Loci Encoding Key Word Binary For Decoding Major System Image For Decoding
A . _ astronaut suits Abs 01 SeaT
B _ . . . bronze busts Bread 1000 DiZZy SoSa
C _ . _ . crystal chandelier Cobra 1010 DoZy DeSi
D _ . . desk Dole 100 oDySSeuS
E . emerald city painting Ed 0 Saw
F . . _ . fountain Flash 0010 ZeuS ToeS
G _ _ . glass case Gore 110 TooTSie
H . . . . harp player statue Heidi 0000 SaSSy SuZy
I . . ivory tusks Ice 00 SauCe
J . _ _ _ jewels Jenny 0111 SaD TeD
K _ . _ kinetic sculpture Knit 101 ToaST
L . _ . . lighthouse Large 0100 SooTy SeuSS
M _ _ marble block half carved statue Mom 11 ToaD
N _ . nests Nye 10 DiCe
O _ _ _ oriental rugs Ozzy 111 DDT
P . _ _ . palace model Pasta 0110 ZiTTy TaZ
Q _ _ . _ quilts Quran 1101 TieDye SaT
R . _ . rock formation Ring 010 CD’S
S . . . stones Sale 000 SoySauCe
T _ tribal masks Ty 1 Tee
U . . _ urn Ugly 001 ZeST
V . . . _ violin Vicar 0001 SauCy Soda
W . _ _ wood carvings Wasp 011 CiTaDel
X _ . . _ x-rays Xraay 1001 DaiSy SiT
Y _ . _ _ yacht model Yukon 1011 TaZe DaD
Z _ _ . . zen garden Zoned 1100 DeaD SeaS

EDIT: Here are some alternate palace locations. ChatGPT came through with loci on a military base which is more fitting thematically with the idea of Morse code for military communications. I changed one or two, but mostly this is from the AI:

airfield
barracks
checkpoint
drill field
explosives
fitness center
gas station
hospital
intelligence center
javelin missile
kevlar armor
landmines
mess hall
night vision goggles
officers club
power station
quarantine facility
radar dishes
shooting range
tanks
uniform closets
volleyball court
water tower
xenon searchlight
y-wings
zip-line training area

1 Like