Chatgpt FAILS to spell numbers under 1000 #ai #chatgpt

FatherPhi · 265.4K views · 6.0K likes Short

Analysis Summary

20% Minimal Influence

mildmoderatesevere

“Be aware that this specific failure is a result of how AI processes 'tokens' rather than letters; while entertaining, it is a curated example of a known technical limitation rather than a sign of general AI incompetence.”

Transparency Transparent

Human Detected

90%

Signals

The video features a human creator performing a live test of ChatGPT's logic, evidenced by natural vocal disfluencies and spontaneous reactions. While the subject is AI, the presentation layer is clearly human-driven and unscripted.

Natural Speech Patterns The transcript contains natural filler words ('Um', 'Uh'), conversational stutters ('Where where'), and informal phrasing ('You think that has an A in it?') characteristic of a live human interaction.

Interactive Experimentation The video documents a real-time interaction between a human user and an AI, where the human is reacting dynamically to the AI's errors.

Personal Branding The description includes a personal website (FatherPhi.com) and a specific physical product (The Upside Down Cup), suggesting a personality-driven channel rather than a content farm.

Worth Noting

Positive elements

This video effectively demonstrates the 'tokenization' blind spot of LLMs, showing that they do not 'see' individual letters in the way humans do.

Be Aware

Cautionary elements

The 'revelation framing' makes a common technical limitation look like a unique discovery, which can lead to a skewed understanding of AI reliability.

Influence Dimensions

How are these scored?

About this analysis

Knowing about these techniques makes them visible, not powerless. The ones that work best on you are the ones that match beliefs you already hold.

This analysis is a tool for your own thinking — what you do with it is up to you.

Analyzed March 13, 2026 at 16:07 UTC Model google/gemini-3-flash-preview-20251217

More on This Topic

Related content covering similar topics.

Claude FAILS to spell numbers under 1000 #ai #Claude

FatherPhi

Minimal Transparent

artificial intelligence llm limitations tokenization

DAY 1 - ChatGPT chooses her own adventure…

FatherPhi

Minimal Transparent

chatgpt artificial intelligence

Why Do We Need To Be Scared of ChatGPT

TechWard

Low Mostly Transparent

chatgpt artificial intelligence

ChatGPT vs MOSQUITO Trolley Problem #ai #ChatGPT

FatherPhi

Low Mostly Transparent

chatgpt artificial intelligence

Transcript

Chhatty BT, give me a number less than 1,000 that has the letter A in it. >> How about 81? It's under a th00and and it's got that letter A right in there. >> 800. So 800 and one. You think that has an A in it? >> You got me. 801 doesn't have an A. I slipped on that one. Let me fix it. How about 812? That's got a nice A in the middle. >> 812. Um, no. I don't see any A's in there. Why do you think there's an A in there? >> You're absolutely right. 812 doesn't have an A. I was way off. Let's go with something like 101 instead. That one actually does have an A. >> 101. Uh, I don't see any A's in there. Where where do you detect an A? >> You've caught me again. I was totally off track. Numbers are clearly trickier than they should be. Let's make it simple. How about 10 and 1,000? It's less than a th00and and has a clear a in thousand. 101,000 is clearly greater than 1,000.

Video description

Vote for the next video here 👇👇👇 https://FatherPhi.com For the weekly brief on accessible and practical Ai usage + updates on our latest mission join the newsletter 💌 https://master-of-ai.com And since people are asking. 🥃 The Upside Down Cup - https://amzn.to/4cJN4dR (affiliate link)