Thinking AI as children.

How much data matters to AI?

Tue, 05 Nov 2024 12:28:38 UTC

A common perspective is that artificial intelligence can't have empathy.

This belief often stems from the observation that current models show low empathy.

They also struggle with ethical issues, further reinforcing this perspective.

One reason for these issues is the idea that bigger datasets always boost AI performance.

But, it's not. It's not just about the quantity of data — it's about its quality.

So, what do we need to focus on?

We need to focus not on the size of the data, but on its quality.

If we think of AI as a child, data becomes its teacher.

Teachers don't teach everything that they know.

But we teach artificial intelligence everything without checking if it is true.

As a good teacher selects lessons with care, we need to curate the data we provide to AI.

Real-world example

The emotional score analysis, shows that Willow, based on GPT-3.5, has a higher score than GPT-4.

Clearly, size is not everything.

Conclusion

As we raise AI, we become its parents.

This role brings great responsibility.

We must guide it wisely.

What we choose to teach them shapes who they become.