Why AI is not as great as the hype

Office-Watch.comLast updated: 20 September 2024

CoPilot,Featured,Microsoft 365,Office 365

All you need is a ‘strawberry’ to see both ChatGPT and Microsoft’s Copilot gets simple things wrong. Or ask which is larger of two numbers. Modern AI systems aren’t as great as the hype and why you definitely need to check anything Copilot, ChatGPT, Google Gemini or Meta etc tells you.

Ask an AI system to count the number of letter ‘r’ in the word ‘strawberry‘. Any five year old can tell you there’s three … but not the much ballyhooed billion-dollar AI systems.

They all insist there’s only 2 letter ‘r’ in the word ‘strawberry!

This is a trivial trick, sure. But it’s an example of how ‘over the top’ the hype is about AI technology in it’s current form.

If AI can’t count letters, why trust it to summarise an important Word document without missing some key details? Or check an Excel table to give the correct analysis?

Anyone who has used AI in the real world quickly finds out that AI is not as great as the marketing promises. Facts have to be checked, text edited, analysis and results confirmed, summaries missing vital points – all things that need one of us mere humans to fix.

Academic proof that AI is wrong

Since we originally wrote this article, the strawberry problem has spread all over the Internet. We’ve obliged to Damien Benveniste for this rigorous academic analysis of the number of letter ‘r’ in strawberry. Surely he’s in contention for the next Ig Nobel awards.

ChatGPT

Copilot

Microsoft Copilot is based on ChatGPT so it’s not surprising that is makes the same mistake.

Google Gemini

Google’s Gemini AI chat system also can’t count.

Meta AI

Same problem with the Facebook / Meta AI

Simple Math can confuse AI

Another example of how AI has some ways to go.

9.11 or 9.9 which is bigger?

A simple question that should not bother a child but AI ties itself in knots before coming up with the wrong answer. This is from Copilot but ChatGPT makes the same mistake.