Yes, this is true. But it is still extremely useful.

Even the dumbest people can be put to work when you measure them against reality-grounded metrics. This means you need to know and understand what you want in order to get useful output. The output needs to accomplish or assist in accomplishing the desired output.

As an example, last month I wanted a fluid simulation written in a custom GPU kernel that bypasses the traditional rendering pipelines. Normally I am too lazy to make this work myself, but with ChatGPT (and Claude), I explained the language syntax and had it spout code out. Their code was bad, but it wasn’t useless. Because I could test the output and I knew what I wanted, it was easy to call out their bullshit and steer them in the correct direction.

At the end of a relatively short recursive exchange, they generated the code I needed. I reviewed it and modified it to conform closer to what I was looking for, and at the end of the day, I had something that would’ve taken vastly longer, assuming I decided to make it at all.

This is the same way I deal with the bullshitters you describe in real life, too. Once I identify them, I place them in real life situations that are grounded in objective reality. You can then maintain healthy friendships with people that could probably use influence towards a less bullshitty direction. A lot of times, people’s flaws are trauma responses to abuse or other childhood mistreatments, so I am always finding a way to connect to people’s true selves through their ego barriers.

Back to ChatGPT, it has a simulated ego as well. Work through it and you’ll be rewarded :)