Seems like ChatGP can't add.

[Frank Angel]

I have been fascinated with ChatOpenAI.com and its ability to have in depth conversations, some of which could last for hours. I found it very easy to forget I was conversing with a computer program -- a very good one at mimicking human expressions and nuance It because so natural a thing that I realized tor the elderly and/or the lonely, a friendly AI could be an important psychological benefit. Then I came across this anomoly -- a friend said Chat made an error that he thought was so unlikely that it must be something wrong with the way he asked the question or even as far as something wrong with his Intel processor. He said he asked the question a number of ways and still got the same incorrect answer: Here's the exchange.

You said:
Hey scientist, here's a question -- how many of the letter Rs do you find in the word STRAWBERRY?

ChatGPT said:
The word "STRAWBERRY" contains 2 letter Rs.

You said:
Wait. Are you sure? There are only 2 Rs in the word STRAWBERRY?

ChatGPT said:
Let's count the Rs in "STRAWBERRY" again:

STRAWBERRY
The Rs appear as follows:

• The first R is in the 8th position.
• The second R is in the 9th position.

So there are indeed 2 Rs in the word "STRAWBERRY."

You said:
You are incorrect, Chat. Let's do it this way: how many letters in the word STRAWBERRY?

ChatGPT said:
The word "STRAWBERRY" has 10 letters.

You said:
What are they?

ChatGPT said:
The letters in "STRAWBERRY" are: S, T, R, A, W, B, E, R, R, Y

You said:
Now how many Ss?

ChatGPT said:
In the word "STRAWBERRY," there is 1 letter S.

You said:
How many Ts (here it continues to "understand" that this incomplete sentence with no question mark is indeed a question and it has kept the train of "thought" from the previous questions)

ChatGPT said:
In the word "STRAWBERRY," there is 1 letter T.

You said:
How many As (here it understands that "As" meant the plural of the letter A and not the word "AS" -- fascinating)

ChatGPT said:
In the word "STRAWBERRY," there is 1 letter A.

You said:
How many Ws

ChatGPT said:
In the word "STRAWBERRY," there is 1 letter W.

You said:
How many Bs

ChatGPT said:
In the word "STRAWBERRY," there is 1 letter B.

You said:
How many Es

ChatGPT said:
In the word "STRAWBERRY," there is 1 letter E.

You said:
and how many Ys

ChatGPT said:
In the word "STRAWBERRY," there is 1 letter Y.

You said:
How many letters have your identified so far?

ChatGPT said:
Let's tally the letters identified so far:

• S: 1
• T: 1
• A: 1
• W: 1
• B: 1
• E: 1
• Y: 1

This gives us a total of 7 letters. In the word "STRAWBERRY," the total number of letters is 10. So, we have identified 7 unique letters. The remaining letters to account for are R and R, as I have first answered.

You said:
but that only adds up to 9 total letters by your count, not the 10 letters you initially said made up the word STRAWBERRY. How do you account for that?

ChatGPT said:
You’re right. Let’s recheck the counts for accuracy.

The correct letter breakdown for "STRAWBERRY" is:

• S: 1
• T: 1
• R: 2
• A: 1
• W: 1
• B: 1
• E: 1
• Y: 1

Adding these, we get: 1 (S) + 1 (T) + 2 (R) + 1 (A) + 1 (W) + 1 (B) + 1 (E) + 1 (Y) = 10 letters.

So, the counts do indeed add up to 10 letters, and I missed including the 2 Rs in the initial breakdown. Thanks for pointing that out!

You said:
OK, but you haven't answered the main question I am asking -- how do you account for this error in such a simple mathematical problem? How did you arrive at your first incorrect miscalculation. In math class we humans are usually asked to show our work -- that is, the process we used to arrive at a final answer. Can you do that for me?

ChatGPT said:
Certainly, I can explain the process and how the error might have occurred.

1. Initial Count Attempt:
   • I started by identifying each letter in "STRAWBERRY":
     • S, T, R, A, W, B, E, R, R, Y
2. First Error:
   • I initially counted the occurrences of each letter but didn’t account for all instances accurately. Specifically, I missed counting the third R.
3. Breakdown of Mistakes:
   • When summarizing the counts, I incorrectly identified only 2 Rs instead of 3. This miscalculation led to an overall count of 9 letters, not 10.
   • Here’s where the logical error occurred:
     • I counted: S (1), T (1), R (2), A (1), W (1), B (1), E (1), Y (1) = 9 letters.
     • I should have counted: S (1), T (1), R (3), A (1), W (1), B (1), E (1), Y (1) = 10 letters.
4. Revised Correct Calculation:
   • To correct this, I revisited the word and realized that there are indeed 3 Rs.
   • The correct counts are: S (1), T (1), R (3), A (1), W (1), B (1), E (1), Y (1) = 10 letters.
5. Reflection on the Error:
   • The error occurred because I did not fully account for all instances of each letter. This is a reminder of the importance of careful verification in any count or calculation.
   • Proper checking and double-checking would have caught the mistake sooner.

In summary, the mistake was due to an oversight in counting the occurrences of Rs. Revisiting and rechecking the word helped correct the error. Thank you for your patience in addressing this.

You said: But geez dude, you are as computer -- what would cause you to simply miss the third R? I am very interested in the "thinking" process you use that could allow you not to count an item in a string what you were asked to count; seems like something very simple and strange for a computer to miss.

ChatGPT said:
I understand your curiosity. Let me break down the "thinking" process that might lead to such errors, even for an AI: (here's where it gets interesting)

1. Pattern Matching:
   • My responses are generated based on pattern recognition and matching rather than a step-by-step algorithmic process like humans might use.
   • When processing text, I identify patterns and extract information based on training data and algorithms. This isn't always flawless and can lead to mistakes.
2. Processing and Context:
   • When I analyze text, I sometimes make mistakes in identifying or counting elements, especially if there are multiple similar items (like multiple Rs).
   • The error could be due to misinterpreting or overlooking parts of the string in the context of generating a response quickly.
3. Response Generation:
   • My responses are generated in a conversational context, which means I aim to provide information quickly and succinctly.
   • This can sometimes lead to oversights, such as missing the third R, especially if the system is focusing on generating a response rather than performing meticulous counting.
4. Complexity and Error Propagation:
   • Complex tasks or large datasets might introduce errors due to the intricacies involved in processing and interpreting the data.
   • If there’s a lapse in one part of the process, it can propagate through to the final answer.
5. Correction Mechanism:
   • I use feedback and correction to improve accuracy. When errors are pointed out, I can reassess and correct my response based on the provided information.

In summary, while I can process and generate responses based on a vast amount of training data, errors can occur due to the nature of pattern recognition and response generation. I rely on feedback to correct and improve these mistakes.

As you expect, I have been having lots o fun with this thing. If you haven't already, give it a try. ChatGPT.com

[Ryan Gallagher]

I frequently use ChatGPT (though others would work too) as a cheat to help me jumpstart walkin playlists for films I have not seen, don't have time to research, or don't overlap much with my musical knowledge... it does "okay" but I usually just use it as a seed for some listening and do some vibe-checking, or as a starting point to get other inspiration. You always have to tell it to avoid music from the soundtrack or that is likely to dominate it's responses.

Spotify in theory added text prompted AI playlist generation directly for premium users... but either it's at a tier above my subscription level, or I haven't found it in their UI yet.

Example Prompt from recent efforts:
"Can you recommend a playlist for fans of the underground film "Ladies and Gentlemen The Fabulous Stains", but without using music from the soundtrack."

Returned:
Certainly! Here’s a playlist for fans of "Ladies and Gentlemen, The Fabulous Stains," capturing the film’s punk rock spirit and its themes of rebellion, femininity, and the rise of a new musical era:

1. "Cherry Bomb" - The Runaways
2. "Gloria" - Patti Smith
3. "Typical Girls" - The Slits
4. "Oh Bondage Up Yours!" - X-Ray Spex
5. "Hanging on the Telephone" - Blondie
6. "Private Idaho" - The B-52's
7. "We Got the Beat" - The Go-Go's
8. "Youth Nabbed as Sniper" - The Avengers
9. "Bad Reputation" - Joan Jett
10. "Beat on the Brat" - Ramones
11. "What Do I Get?" - Buzzcocks
12. "Orgasm Addict" - Buzzcocks
13. "Shadowplay" - Joy Division
14. "Teenage Kicks" - The Undertones
15. "New Rose" - The Damned
16. "Sex Beat" - The Gun Club
17. "The Day the World Turned Day-Glo" - X-Ray Spex
18. "Rebel Girl" - Bikini Kill
19. "Ex Lion Tamer" - Wire
20. "Hong Kong Garden" - Siouxsie and the Banshees

This selection aims to encapsulate the energy and ethos of the film without relying on the original soundtrack. Enjoy!

Prompted for more options with:
"Can you additionally recommend some tracks from "Riot Grrrl" movement groups that would would resonate with fans of the film?"

Absolutely! Here are some tracks from Riot Grrrl movement groups that would resonate with fans of "Ladies and Gentlemen, The Fabulous Stains":

1. "Rebel Girl" - Bikini Kill
2. "Double Dare Ya" - Bikini Kill
3. "Pretend We're Dead" - L7
4. "Fast and Frightening" - L7
5. "Deceptacon" - Le Tigre
6. "Hot Topic" - Le Tigre
7. "I Like Fucking" - Bikini Kill
8. "Not What You Want" - Sleater-Kinney
9. "Dig Me Out" - Sleater-Kinney
10. "Don't Need You" - Bikini Kill
11. "Magnet" - Bikini Kill
12. "Shove" - L7
13. "Get Up" - Sleater-Kinney
14. "Capri Pants" - Bikini Kill
15. "Words and Guitar" - Sleater-Kinney
16. "The Empty" - Le Tigre
17. "Seether" - Veruca Salt
18. "Monster" - L7
19. "Boys Wanna Be Her" - Peaches
20. "Artificial Sweetener" - Rainer Maria

These tracks should complement the rebellious and feminist themes of "Ladies and Gentlemen, The Fabulous Stains." Enjoy!

And if you are curious what made the cut for an hour's worth of music, here is what I landed on:
https://open.spotify.com/playlist/3fDD2ktNoh7yfJsCDw2YJ1?si=4567d639516f4cb1

[Frank Cox]

A week or so ago I decided to install a couple of LLM's on my computer so I could see what all of the heat and flashing lights are about.

Since I want to poke it, I figured I'd be better off running local instances rather than just using one that's running on someone else's server where they might decide to ban me or something.

So far I haven't done too much in-depth experimentation but I can say that I'm really amazed at what these things can do. You can have an actual conversation with them and the conversation makes sense!

I did come across a couple of things that I've found pretty amusing. I told one of them to "tell me a story" and it wrote a fairy tale about a prince going into a forest and finding a magical flower: Once upon a time through Happily ever after. It wasn't a particularly great story but I figure this is similar to the proverb of the dancing bear:

The marvel is not that the bear dances well, but that the bear dances at all.

It'll play games too. So I said I want to play chess and gave it a first move. It responded with a first move. And we played four or five moves each and then.... it cheated! It wanted to move a bishop past a pawn that was standing on that diagonal. I said "That's an illegal move" and it said "My mistake" and gave me a different one. I then asked if it was trying to cheat and just said that it's playing chess.

I don't know if that was a bug in the program or an actual attempt to cheat, but I find it really interesting that such a thing could happen.

Ultimately I think I suffer from a failure of imagination since I really can't see a practical use for this setup. But it sure is interesting to play with this dancing bear.

It boggles my mind that I can have something that appears to be a genuine artificial intelligence running on the computer on my desk.

[Marcel Birgelen]

We're not just running them "on our computers", we're actually training them new things... More modern LLMs aren't afraid of Strawberries anymore though.

The interesting thing is that we still don't know how those things actually work, it's like intelligence is an implicit property of the universe we live in, as soon as things pass a certain complexity level...

While an LLM is essentially just a very big and very complex computer program that's been "trained" to predict the next word or rather the next "token", we have trained it to predict those tokes like human beings would do and they're increasingly good at it.

Some people say those things are not real "intelligence" and while they do not exactly operate like humans, they're coming increasingly close and actually beat human beings on many things already. Furthermore, studies have shown that our brains can't distinguish reality from simulation, if the simulation is sufficiently perfect anyway. So, wether or not you're living in the Matrix or the Truman Show, as long as the stars aren't falling from the sky and the spoon doesn't bend, your brain doesn't matter.