@cwebber you don't by any chance have a magic want that instead replaces our socioeconomic system do you? 😅

@cwebber if you could wait for the bubble to go down and ram and storage and GPU to go down to a reasonable price first, it would be great :)

@cwebber Personally I try to avoid using AI. But I often think about it because I am surrounded by AI enthusiasts who almost convinced me that I need it to be somehow able to make a living as a lawyer (ok in a perfect work we would not need lawyers but this is another topic). But after all these discussions, the above arguments don't convince me anymore that AI is inherently bad. What worries me more, is that people lose the ability to think for themselves. I don't know if this can be fixed.

@cwebber I think that the points make sense even in actual socioeconomic system. ActivityPub worked, overall.

However, the wand would just make era of "slave AI" sooner to come, which makes me sad. Anyway, I am not sure if we would be ever ready to avoid that era...

@cwebber

I mean, you skipped the whole "totally unreliable" part.

But even including "gives you exactly correct and accurate information with context every time", what you're really asking is "what if I could change LLMs *entirely* into a new technology that didn't have any of the failures and externalities of LLMs and wasn't just a sophisticated Markov chain but actual AI?"

and sure why not 🤷‍♂️

@cwebber No mention of compensating the people whose work these bastard genAI companies have stolen without even a hint of permission or recompense?

@cwebber I have been playing with existing models, and I keep coming back to the same thoughts: Its not the tool in the end, it's the society and it's frame which produces the outcomes of the tool.

I used the term "Fully Load AI" in conversation set the frame when I speaking of AI to non technical people. I fear humans are not mature enough (in the 24th century Star Trek way) to handle any of it; but also information has its own velocity and we are rapidly circling the drain.

Making LLMs 30% more efficient today has real benefits, but also negative externalities.

@cwebber If the “mores” and “lesses” are of adequate degree to say “Yes,” corporate capitalism would by necessity change substantially. It might not die, but actually working AI that is available to anyone, isn’t a moral disaster, and is capable of consistently well-explained accurate output would cripple many of the tools used by the capital class to maintain and enhance their wealth.

I doubt that any LLM will ever get to the point of justifying its own existence in human-based terms.

@cwebber

You are describing a human co-worker here, so I'd say there is no need to reinvent that wheel.

@cwebber Since for some (if not most) of these there should be absolutes, I don't see it improving things. I would say that the current foundation of models doesn't allow for any of this. For reference, I read the magic wand as improvements on the current paradigm, not a new one.

@cwebber
Make things slightly better or leave as is? There doesn't seem to be an explicitly stated downside, so what is there to be conflicted about?

The awaited revolution isn't likely to hang on the precise scale of BigAIs copyright violation...

@cwebber Votes are more split than i expected, which means it's probably an interesting question!

I voted yes, because not mentioning an alternative want of making it go away (or changing the present socioeconomic system), i think the question becomes moot, better is better, these things are real problem worth improving.

I don't think we can really imagine going back to how things were before these systems were built (and neither pretend things were void of many of the issues we have with AI systems), even if the bubble pops tomorrow and frontier models' real operating cost become unbearable for most people, it would force people to adjust their usage, but it wouldn't be "back to before", it would eliminate the usage that don't make economical sense, but not all of it.

I don't see how any of these particular improvements would make their usage worse. Maybe it's a failure of my imagination.

@cwebber "More environmentally friendly" is nonsense in the current system, anything more energy efficient would simply be used more often until it reaches the same or greater GHG emissions, due to the Jevons paradox. (That's likely to be a problem in alternate systems as well, in fact it would be the first question I'd have about any alternative.)

If "energy efficient" is not what you mean by more environmentally friendly, you need to get a lot more specific.