Cloud models are bad but local models are not good.

[mcc@mastodon.social]

Cloud models are bad but local models are not good.

[mcc@mastodon.social]

Software that exfiltrates local state to a second computer is a bad thing. Software that uses neural network models is also a bad thing. Thus, cloud models are two bad things, and local models are one bad thing.

Is one bad thing better than two bad things? What kind of question even is that? If the only choice computers give us is how *many* bad things to tolerate, then why do we have computers?

[kouhai@social.treehouse.systems]

@mcc I disagree that neural networks are the distinction; it is rather the accessibility and centralization which matters.

[rotopenguin@mastodon.social]

@mcc this shit sandwich is awful, but at least the bread is fresh

[mcc@mastodon.social]

@kouhai I just don't think they're even the same discussion. Microsoft was moving things willy nilly to undisclosed cloud services before "AI" and they will continue doing so if the coming crash forces them to stop doing cloud models. "AI" is not the cause of cloud offloading. It's much more likely that the reverse is true, AI-branded tech is being adopted *exclusively because* it gives vendors an excuse to shove a cloud API in there.

[mcc@mastodon.social]

Okay yes yes by rejecting neural networks I'm rejecting all the wonderful old technology which was based on neutral networks/machine learning, like motion smoothing TVs, and price-fixing algorithms for landlords, and transformer-based predictive text keyboards which are slower and make weirder errors than the trigraph systems they replaced. Whateverrrrr

[julian@activitypub.space]

@mcc@mastodon.social painting neural networks as bad using such broad strokes seems short sighted... at least from a data privacy standpoint the local models are contained and I'm causing my own energy bills to go up?