I don't need a human to be particularly good at anything to like them. Maybe that is how you work, but the idea is just misplaced to me.

Relative to ever rising hw requirements of apps they obviously get slower. That is why I personally buy new phones.

This is potential tyranny aimed at Anthropic, specifically.

What? Anthropic is not a TikTok sensation. It's a business tool. Businesses need to know their tools work reliably.

When you are situated in a banana republic and the chief banana is out to get you (and demonstrates that they can and will on a whim) that is not great hype but a potential death sentence for you as a service provider.

You are one degree away from becoming forever branded as unusable. (Theoretically until people trust that a sane administration is in control again, but that might as well be forever on current AI timelines, given how much cashflow you need just to keep going)

Think WordPress installations: Depending on how it's done you can either tell at a glance (probably ~90% of WP installations at some points in time) or you have no clue until you look at the html source.

Of course, when given the option to not do it properly is always alluring and then you can tell.

Everyday AI writing was not a thing with GPT 3.5. It happened more around GPT 4o. And now some people are entirely comfortable with using AI writing and not even trying to hide it (while, I would agree, it's obviously still fairly garbage and easily identifiable, which helps with triggering strong averse reactions).

However the models are getting better at everything, including writing for the past years. Why would that stop now? It's reasonable to assume that the makers also know about bad writing, dislike it, and thus the models will get trained to get even better at it.

Eventually how will you be able to tell? You won't. You can't. And that goes for the rest of us. And I suspect everything will just feel somewhat nicer.

Because why not? AI will simply on average be nicer to talk to than most humans, with clearer thinking and better arguments, less contradictions, and easier to comprehend.

I don't know how humans could compete with that (but it also does not seem all that horrible, given that it will be available to every human.)

This is not to say that this idea is uncomplicated or comfortable, in different ways. Just that I think it's true and that it might even be good.

Let's just assume it was "only" that?

It's unreasonable to assume they are aiming to upset people who are just giving them money in the way they want. It makes no business sense, for any company. So that has to be a byproduct.

Model training is one of the more expensive undertakings in the world right now and distilling models from competitors against the TOS is apparently something that is going on for very little money. Why would they not "just" try to take measures against that?

In isolation it's not, but I think it's somewhat lazy to not talk about what they are trying to guard against, when we are supposedly giving the absolute maximum benefit of doubt.

Are we just concluding "their concerns were never real"? Because that probably runs counter the things that they have been observing and concluding.

Anyhow, my research summary: Individual humans are so fucking expensive to train and upkeep (and this includes everything from before womb, where another human already limits their ability to work) You retain ~zero knowledge after death and start all over again for another measly 15 years of effective, productive work. Model training/r&d in relation, when deployed and used at scale, rounds to zero, even with the current retraining regime.

*Of course, the ratio can go to negative infinite if one assumes that models are doing 0 useful work currently and never will

On the margins, suppose the prompt is literally: "Build a feature complete, high polish Facebook clone". Facebook is complex but likely not super complicated tech, and still I would assume that (after having burned through a substantial amount of tokens) you would find substantial enough differences in the outcomes between different models on that prompt on various fronts.

The above ask is obviously not useful, but what's preventing you from taking on bigger chunks until you approach the limit? At some point you would hit a boundary, where the diff will be obvious.