pouët.net

As this topic and the surrounding controversy seems to frequently infest more or less unrelated discussions, here's a thread to spill all your praise or hatred or actual good-faith discussion of pros and cons into.

This is fully expected to go residue soon, but that's okay!

Here's a point that rarely seems to come up: asymmetry of complexity.

While training an AI is extremely resource-demanding, using a trained model for inference is comparatively cheap. Cheap compared to training, but still rather expensive.
Not more expensive than running a PC demo, though, most likely.

However, the sheer amount of cloud-AI inference requests adds up to a multitude of the resources that the limited training phase used, and keeps adding up.

Locally-run special-purpose AI on your computer, maybe... may still not be all that bad an idea.

my point still stands: humanity is not ready for this technology.

..and yes, ideally we should limit this to locally-run AI.

but Pandora's box has already been opened, sigh.

Yeah, I definitely don't think those models are something most people should use, there are much more energy efficient methods to convey information. Probably would make sense that some humans would use language models to write better static text. But that doesn't seem to be the case how everybody is stuffing "natural language user interfaces" everywhere.

It's somehow fitting that Sam Altman was a cryptobro before he became an ai-bro.

Genau. When we talk about “AI art machines”: Supercomplex, high tech, centralized, big money owned machines trained with ethically dubious principles replacing simple pencil manufacture. I can’t believe someone calls this “democratization”. Superfluous complexity and superfluous level of abstraction. Removing the artist one more step from the actual art making.

cont to my previous post: And of course i'm basing is on the axiom that climate change is causing this planet to be largely inhabitable for a lot of humans, and that is caused by the excessive energy use that our socities have been incresingly doing. So while AI is probably far from the worst offender, it certainly doesn't help.

Also people seem to be mesmerized by the human-like talking companion, especially if they are lonely. Then their world view becomes more and more skewed. Internet is already a toxic place.

And also since it's just based on statistical modelling of a large amount of human-created material, it's also questionable to me how an copyright law can ever be meaningfully enforced after the corporations are selling products based on this data directly.

Otherwise is just an another tool.

Sure. If it doesn't burn the planet or is trained on copyrighted material.

We should steal an AI, crack it, put some nice intro and maybe trainer to it, then make it work for us.

Facebook's first Llama model was originally leaked by some researcher and distributed via bittorrent. Cracktro was missing though :-)

Interesting. This thread was residued for no apparent reason other than the topic itself, thread so far devoid of any personal attacks and the like.

Anyways, may it serve its purpose! =)

just what pouet needs... another AI thread!
(sarcasm-sign for the Sheldons among you)

It needs a good-faith ai thread, with no posts like that