YourNetworkIsHaunted

I mean that's the fundamental problem, right? No matter how much better it gets, the things it's able to do aren't really anything people need or want. Like, if I'm going to a website looking for information it's largely because I don't want to deal with asking somebody for the answer. Even a flawless chatbot that can always provide the information I need - something that is far beyond the state of the art and possibly beyond some fundamental limitation of the LLM structure - wouldn't actually be preferable to just navigating a smooth and well-structured site.

Also I can guarantee that "her" system prompt includes the phrases "truth-seeking" "fun loving" and "kinda hot".

You know, I think that even the brick comparison still favors the traditional dictionary. After all, if you wanted to use it for traditional brick purposes like holding doors, weighing down papers, or throwing at [redacted], you would need to invest in a bunch of GPUS and burn them out generating a bunch of worthless slop, spending God only knows how much water, energy, and time. I want to save the rainforest as much as the next guy, but I think it's obvious that the paper brick is far more cost-effective for both the user and the environment. And provided you can clean the dirt and blood off afterwards, you can even still use it as a dictionary.

I imagine it helps with the slightly plastic look that skin still gets too

Shaka, when the walls fell?

Sadly no. Though thank you for making my life immeasurably worse by sharing that.

God save me from arguing with HPMoR readers.

I need you to understand how nearly I added myself to even more watch lists because of that analogy. Avalanche!

My name is Rocco Pa'perclip Basilico Yudkowski Way...

It is kinda fun to think about the counterfactual world where this shit had worked as expected on the first try but for obvious reasons didn't hold up in copyright court, making the entire catalogue used for training functionally public domain.

Don't mock me dammit, I'm allowed to dream.

Technocapital identifying a real problem only to frame it in a way that puts the blame on users/workers rather than themselves? Shocking development. Usually they can't come up with half that good a name for it.

This is a fascinating case study in the democracy vs technocrat debate, where the educated, well-paid consultant class is too far up their own ideological asshole to recognize that their GenAI chatbot doesn't work as advertised, and that lack of sanity is poisoning the well and killing discussions over how to use the power of machine learning well. But rather than back off the chatbot, even just to "let it cook" (read: let Saltman and friends continue to play with it in arenas that don't have geopolitically relevant stakes) they're doubling down to make this a pure messaging issue which I expect is only going to make the problem worse. Like, people aren't unaware of GenAI at this point, and education on prompting it right is just a distraction from the fundamental problem.