Hey Reader!

The internet is filled with ChatGPT failures.

From the inability to count the R's in the word strawberry:

To being unable to answer simple puzzles correctly:

But the bottomline is that this goes way beyond simple things.

Go to ChatGPT and type:

And you will get something like this:

Notice the 10:10 time.

ChatGPT almost always generates images at 10:10.

Because that is most of the training data.

Let's try a different one.

Identical, right?

ChatGPT’s internal clock is stuck on Groundhog Day.

And this is a simple thing.

Clocks at different times are not part of the training data and, therefore, we can almost say that the inference capability of AI is non-existent.

Do you want to see something even funnier?

I went ahead and asked the current state of the art OpenAI model to recreate this image and loop the output 100 times.

If you are interested in the code, you can find it here:

Before you scroll down, tell me what you think the output is:

I will admit, I was not ready for this.

But before I show it to you, here is the point of the email

AI is a weapon.

Weapons alone are useless.

Your goal is to become a master of wielding the weapon AND being experient enough in all situations.

So, yeah, AI will never replace me.

Here is the gif:

If the model can’t make me hotter after 100 tries, my gf is safe.

And what about you, are you still afraid of being replaced?

To being irreplaceable,

Diogo