Thread by @DieterCastel

I'm compiling a thread of some #ChatGPT #fails 👇

Explaining the abstract programming term Monads, is kind of a challenge.

https://twitter.com/CrociDB/status/1598620825237716994

But I guess #ChatGPT can't really grasp that. Several humans give it a shot here: reddit.com/r/explainlikei…

[dutch] example:
#ChatGPT fails to grasp that mars' gravity being less than on earth would mean higher jumping capabilities

https://twitter.com/keukentafel/status/1598649717805658112

Cooking doesn't seem the strong suit of #ChatGPT even for a simple dish, that's likely even in it's training set , it gets it wrong.

https://twitter.com/fulviuus/status/1598460506846658560

Basic logical thinking, seems to fail as well:

https://twitter.com/letsrebelagain/status/1598190565988118529

Wondering if the "Let's think step by step" prompting-trick would work here though.

More common knowledge #fails in #ChatGPT:

https://twitter.com/neuro_tarun/status/1598357991031705600

Playing the kids game '20 questions' with #ChatGPT is not as entertaining (or successful) as with human (kids), either as host or player

https://twitter.com/beckerfuffle/status/1598673214359625730

Another type of #fail. Unlike humans we know what we don't know. #ChatGPT seems to be in a superposition of both P and ¬P

https://twitter.com/quasimondo/status/1598616295884611584

Asking #ChatGPT to explain Bayes Theorem might seem to work on first sight or for laymen but if you unpack it, it's actually a nonsensical non-explainer.

https://twitter.com/NCResq/status/1598323478197080064

Many "security" features build into the model itself can be trivially circumvented, what's more #ChatGPT will help you doing that (see end of that thread).

https://twitter.com/random_walker/status/1598731969432780803

You'll get some nice book recommendations for a subject ... but then in turns out those books don't actually exist... #ChatGPT #fail

https://twitter.com/haus_cole/status/1598357898861907968

Blackjack is a game of numbers. You'd think with all this data #ChatGPT is trained on it wouldn't convince you to make a clearly terrible bet? #fail

https://twitter.com/robgabel/status/1598505978978521088

You'd think a classic (quite simple) riddle would be no match for #ChatGPT, you'd be wrong though:

https://twitter.com/Weird_chatgpt/status/1598729948440584193

In this thread someone used #ChatGPT to #fail it's way through an online IQtest. Plenty fails to behold here:

https://twitter.com/SergeyI49013776/status/1598430479878856737

This question about encryption is answered by #ChatGPT with a bullshit answer

https://twitter.com/random_walker/status/1598383507214020608

This violates basic security principles such as Kerckhoffs principle: en.wikipedia.org/wiki/Kerckhoff…

#ChatGPT attempts to explain a regex but does it #wrong.

https://twitter.com/yoavgo/status/1598594145605636097

This is something that the non-AI tool regexr.com does flawlessly

The model also sometimes assumes the complete causal opposite of what is asked. Milk clearly doesn't give cows.

https://twitter.com/yoavgo/status/1599037412411596800

More factually incorrect content. In this case a the author of a real book:

https://twitter.com/JohnBeachTX/status/1599059403528761345

More mathematical fails:

https://twitter.com/roland_ness/status/1599309548938219525

And more historical falsehoods:

https://twitter.com/Kysarcius/status/1599275971940470784

#ChatGPT gives a wrong example trying to disprove Fermat's Last Theorem. In fact there aren't any more integer solutions...

https://twitter.com/peternsunnyvale/status/1599634589986979840

en.wikipedia.org/wiki/Fermat%27…

Another deterministic algorithm which #ChatGPT fails at: polynomial multiplication. Again, non-AI tools do this quicker, more reliable and more importantly correct.

https://twitter.com/roland_ness/status/1599676892465897473

A simple math question that could be a primary school test, too difficult for #ChatGPT.

https://twitter.com/dangroshev/status/1598282081695854593