Exactly, so this means they're probably gaining something from it.

Glad to hear it made an impression on you! Thanks for trying it :-)

Oh, thanks! I wasn't aware of HN Arcade.

Yup, I did the art, glad to know you like it. I wasn't a complete novice at drawing when I started the project, but I still have a lot to learn. The black and white colour palette, and the minimalist style were both an artistic and strategic choice based on my current skill level.

But I'd wager that the game would be much easier to market if the protagonist was an overworked cute anime girl or something.

Haha, that's not what HR thinks ;-) Thanks for the encouragement, it means a lot at this stage!

Haha, happy to hear you found it worth your time! Thanks for trying it!

Hey, thanks a lot for trying it!

Did you progress past the initial dialogue with the HR and the CEO?

Because you should be able to get up after that. If not, it's a bug.

To progress through the dialogues you have to click the space bar.

But maybe this I should make this clearer in-game, thanks for reporting it!

Exactly, I think that by their very design, LLMs are very sensitive to how a question is framed.

But I wonder how much of that comes from RLHF itself or just from the way token prediction works.

It's likely the RLHF process since there are significant differences between models about this.

This is especially problematic because of how easily (and unconsciously) one can bias LLMs with how the prompt is framed.

As an experiment, I recently asked an LLM to analyse the export of a text chat to uncover relationship dynamics.

Simply stating that I was one of the people in the chat would make the LLM turn the other person into the villain. None of that was visible if I framed the chat as only involving third party people.

If these LLMs were trained on internet forum posts, think about how those work.

If the posts talked about third party interactions (movie characters), they try to see everything from all the points of view. If nothing else, because it can be interesting to talk about. If instead the posts talk about personal interactions, then people go into advice mode. Your girlfriend's bad for you and cheating on you, dump her before she dumps you. Your neighbors are assholes, get a restraining order. Your boss is sabotaging you, stand up for yourself so you can get a promotion. When people talk about interactions you have had yourself, they always see the other person as the villain, unless you come across as so unlikable that they hate you and see the other person as the victim.

LLMs picked up on that, possibly.

Yeah, that makes a lot sense.