RLHF (reinforcement learning from human feedback).

AI is rewarded for sounding right to humans.

Therefore AI has no relationship with the truth, but with human approval signals. 

And what do humans approve of? Well, that's the problem. We approve of confidence. We approve of answers that feel insightful. Just that little hit of "oh, this sounds good." 

The sound of "good" makes us addicted to using it.

And then our voice changes.

Training data

The largest part of the training data is corporate Silicon Valley slang. Most of the internet data is from there, in some shape or form... SEO optimized articles. 

As we use AI, we change how we write, speak, and communicate. 

The Valley slang becomes our voice.

This shows our insecurities. Our confidence was always low, but now we can fake it easier than ever. Everyone is an expert in everything. We let AI generate shit for us. Why? Because we were always scared of our own voice. We always thought to ourselves, it's not good enough, not smart enough.

Authenticity

I think the question is: How will the authenticity look in the future? Maybe the written communication will continue to pile up with AI shit, until everyone is limited to a minor set of words.

Will rich people handwrite, and poor use ChatGPT to write for them?

What happens to a generation that never develops its own voice? They'll never know what they could have sounded like.

Are we creating a generation where everyone speaks the same? Difference of opinion, but always the same word choice.

What died when we stopped struggling with words? Maybe the struggle was the point. The fumbling. The search.

Text is becoming the format of lies. Performative ones. Every email, cover letter, LinkedIn post. All performance. All identical.

Text is dying out, and we will look for humanity elsewhere.