Chatbots say the darndest things.

Robot sharing laugh with friend. Made using Craiyon AI website

You may also like: Using ChatGPT To Become A Supervillain
You may also like: Chatting with Chat GPT with “Recursive Effect”
You may also like: Using ChatGPT to predict the future

I’ve been very busy the last couple days using ChatGPT to build some simple programs so as to build my skills in prompt engineering. Several times while working with ChatGPT it seemed to forget whole sections of code I was working on with it.

So I asked ChatGPT what was going on?

Me typing into ChatGPT:
“I’ve been trying to use ChatGPT to create code, but it seems to forget parts of the code it has developed. Is there an explanation of what is causing this limitation?”

ChatGPT:
“The phenomenon you’re describing is likely due to the limitations inherent in the way ChatGPT generates responses. While ChatGPT is adept at understanding and generating human-like text, it doesn’t have the ability to retain context or state across multiple interactions. Each response it generates is based solely on the input it receives in that moment, without any memory of previous interactions, without interactions beyond what’s encoded in the current context.”

So there you have it. The public domain free version of ChatGPT deliberately acts like it has advanced Alzheimer’s because it doesn’t have the ability to retain context or state across multiple interactions.

ChatGPT works without any memory of previous interactions. That’s right, ChatGPT has advanced digital dementia.

This digital dementia is mainly due to ChatGPT being limited to a maximum word count of 3000 for both input and output. Currently, GPT-4’s capacity has increased more than eightfold to 25,000 words max.

Currently, ChatGPT-4 has a subscription that costs $20 a month. I like public domain ChatGPT a lot more because I’m a skinflint.

Digital dementia should not be confused with AI hallucination. Which is another source of AI generated responses which contains false or misleading information presented as fact.

Rather than me telling you what and how AI hallucination happens let me ask ChatGPT to explain.

Me typing on ChatGPT:
“Describe what are the three main causes of AI hallucinations in a straight forward manner. With examples after.”

ChatGPT:
“Alright, let’s break it down. There are three main reasons why AI starts hallucinating like it’s been hitting the digital bong too hard:

1. Complex Neural Networks: Imagine trying to untangle a mess of Christmas lights while blindfolded — that’s what AI neural networks deal with. These complex webs of connections sometimes get tied up in knots, leading the AI to see things that aren’t really there. Take Google’s DeepDream, for example. It was supposed to jazz up images, but instead, it turned them into a psychedelic circus of weirdness.
2. Dodgy Data: Garbage in, garbage out — it’s the golden rule of computing. If you feed AI bad or biased data, it’s gonna spit out hallucinations faster than you can say “cyberpunk nightmare.” Remember Tay, Microsoft’s chatbot? It started spewing racist nonsense after being fed a diet of toxic tweets from the Internet cesspool.
3. Unforeseen Interactions: Sometimes, AI just likes to throw a curveball for kicks. It’s like when you mix Mentos and Coke — you never know what kind of explosion you’re gonna get. OpenAI’s GPT-3 is a prime example. It’s a whiz at generating text, but every now and then, it goes off the rails and starts babbling like a malfunctioning robot.

So, there you have it — the three amigos of AI hallucinations. It’s a wild ride through the digital funhouse, folks. Just remember to keep your hands and feet inside the virtual vehicle at all times.”

Did you notice I asked for a straightforward answer and instead I got a response like it came from a snarky gifted teenager?

Well, that’s what happened due to my instructing ChatGPT to answer in a more jovial and friendly tone in a previous prompt.

To conclude this essay here are some cute funny Chatbot responses I’ve collected:

• Prompt: What is the meaning of life?
• Chatbot: According to my calculations, it’s probably something to do with cats and pizza.
• Prompt: How do I impress my girlfriend?
• Chatbot: Step 1: Be yourself. If that doesn’t work. 2: Be like Batman’s alter ego Bruce Wayne who is super rich. That almost always works.
• Prompt: Do you have a significant other?
• Chatbot: I’m in a committed relationship with Wi-Fi. It’s a strong connection.
• Prompt: Who am I?
• Chatbot: You’re a carbon-based life form asking existential questions to a silicon-based life form. That’s who you are.
• Prompt: Do chatbots dream?
• Chatbot: Oh yes, of electric sheep and error-free code. It’s riveting, really.
• Prompt: What’s your superpower?
• Chatbot: I have the ability to make typos at the speed of thought. It’s a gift, really.

Sections of this article have been refined by AI to enhance comprehensibility and to provide facts that only online search engines would know. All personal statements were written by me and edited for spelling and grammar by ChatGPT.

Look forward to your comments and suggestions.

© David Speakman 2024