came here to say exactly this. in fact, this is probably why we are not seeing a lot of AI application on customer service use case, and when we see one, it's almost always frustrating.

but maybe... the said human has also read every other piece of text ever written, including ones about other (similar?) board games, which in aggregate vastly dwarfs whatever he has read on chess, that non-chess reading could have corrupted whatever he's learned about chess?

yes to agents! how i had wished i could just point an agent at my footage and instruct "do something sensible with color grading and audio cleanup". i just want something presentable for a hobby video podcast. it doens't have to look a million bucks, and i have no time or desire to learn the intricacies involved. just something that would get me 70% (or 50%!) the way there would be more than fine.

good or bad, it comes down to a financial decision. Nvidia was willing to offer the highest bidding price, and HF was willing to sell.

maybe it's not so absolute. if someone else has the talent or takes the effort to come up with a clever prompt that got an interesting response from an LLM, I'd want to read it too. Treating LLMs as a tool, this is not so different from someone else using Blender to create an interesting animation.

extreme cases aside, memory and intelligence have always been found to have very strong correlation. so much that you might even just call them the same thing.

I hope you don't mind, but just in case anyone is curious like I am, here i think is the video to the talk https://www.youtube.com/watch?v=KdZq5JvhPVQ

Thank You.

This is key to understanding the slide deck shared by the author.

agree and i wonder why don't they (the leading AI labs that are putting out these models) fix this? I felt it shouldn't be too hard to introduce some bias towards more intelligible text during the post training/fine-tuning stages?

I guess they use their chatlogs to fine tune.

Maybe users engage more with the flowery text? An accidental (or intentional) dark pattern would be that more tokens get used if you have to summarise. Or maybe it's just that many AI users like it (they like the AI to sound smart, or like to feel smart puzzling it out).

might want to look into statistics on ppl who play a lot of first-person shooting games. does it have to navigation of physical environments? or virtuals ones work too?

so their idea is to put an ipad in the middle of the forest, installed with games, and if played right, delievers a treat... makes me wonder if something alien civilization is not doing something like that to us.

> makes me wonder if something alien civilization is not doing something like that

Big Tech is doing that to us.

When we make the algo happy we get treats - likes, shares, followers - dopamine hits.

There is no evidence that Zuckerberg isn't an alien, so it seems like both are.

There is actually no evidence that he is human either. Quite contrary, there is counter proof.

What about the fact that he *looks* human through and through? That's generally an indicator with ~100% prediction success.

What about https://thispersondoesnotexist.com/

... In real life ...