Is ChatGPT4o omnipotent?

ChatGPT4o can read human emotions and laugh at jokes. Entertaining as it sounds, there are risks.

Transcript

00:00ChatGPT 4.0 is more human than ever.

00:03It can communicate emotionally and see its surroundings.

00:06The O stands for Omni, as in omnipotent.

00:09The model can work with text, audio and video,

00:12making the interaction more natural.

00:14Take a look.

00:16And I wrote one last thing I'd love if you could take a look at.

00:19Of course, I'd love to see what you wrote.

00:22Show it to me whenever you're ready.

00:24Okay, so this is what I wrote down. What do you see?

00:26Oh, I see. I love ChatGPT. That's so sweet of you.

00:33According to its makers, GPT 4.0 cannot only express emotions.

00:38It can recognize your emotions, too.

00:40It looks like you're feeling pretty happy and cheerful.

00:43AI and systems learn to recognize emotions by being trained on corresponding data.

00:49So, for example, by analyzing text that links a smile to happiness,

00:54that links tears to sadness.

00:56And AI is trained in an optimization process to recreate this data.

01:05But what happens when AI adopts more human-like traits?

01:09Could the model become indistinguishable from actual humans?

01:13We tend to be very trusting of computers' decisions,

01:16even if they contradict our own beliefs.

01:19That's according to research from as early as 1999.

01:23And more recent studies suggest the same.

01:26Chatbots are already convincing.

01:28But if they speak to us like a human being would, they might be even more so.

01:33And that could be a problem because...

01:35AI can lie.

01:37It can invent facts it believes to be true.

01:40This is called hallucination.

01:42Tests show that AI playing poker or video games

01:45have learned to bluff to produce a favorable outcome.

01:50These systems are known to be lying or to making up things.

01:53And if something is very convincing, empathetic in making things up,

01:57just by natural behavior, I'm way more inclined to believe it

02:01than if it were dry, objective text or reasoning.

02:08Then I can much easier assess, well, this is sort of nonsense or not.