Squid🦑

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

The entire claim that doing this were somehow incompatible with those laws just... doesn't follow out of an already incorrect argument about how the models even work, or how detection of them works. Its just one layer of wrong on another layer of wrong on another layer of wrong.

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

And hey, we happen to provably know that neural networks are in fact capable universal function approximation, if just made big enough. If you make it big enough, and by extension stupidly expensive to run, it can do the task. The cost is just _slightly_ non linear, though.

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

Sure, most of those companies are just in for a quick buck and don't give a fuck if their tools are accurate, but to claim that some law of nature were violated by such a machine existing is just nonsense.

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

In general, things the LLM doesn't need to do very well, or doesn't need to focus on all that much. And well, the people training these models do actually have a self interest in keeping them distinguishable. Feeding older AI slop is, in fact, not great for making a newer slopinator.

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

The actual technical question with LLM detection feasibility now is if those profiles share certain properties that human profiles do not. The answer is, probably yes for old models, and absolutely for most new ones, because AI labs intentionally fine tune them for it.

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

One example is that you can analyze the usage of location specific slang or terminology. People have accents, and not just in their speech! As a result, with enough text, it is very much possible to specifically identify the author of something!

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

Also, the thing is that, a text or any other thing, that has all the average values of a large diverse population, is very strange! A text that is statistically closest to the overall dataset average of all human text, is not very human look at all!

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

And if we assume that was possible, and that such a detector could be used to train better models, which in turn would lead to better detectors, that would not break any laws of information theory of thermodynamics either! That would still require immersive amounts of compute!

That is not how any of this works. This is not even close to why its hard to make a machine like that (and very much possible to do so, in theory).

The model does not produce statistically indistinguishable from human outputs, but the outputs that satisfy its training rewards system most strongly. Those are very different things. And the rest of the argument is less logically.

when you suddenly realise you didnt eat properly in like three days and that maybeee thats why you are doing badly..... anyway food time!!

This is some of the weirdest, most just hate fueled things I've seen in a while. Like, 'DefendingAiArt' by actively threatening to do... what? Train a model on it just to be a awful person? Just, what does this person gain from this?

Also, that implied idea that they somehow think this is beneficial to their cause of normalizing 'AI art.' Being an awful person is very rarely good for convincing people. (Something that everyone needs a reminder on though from time to time, for what its worth.)

This is some of the weirdest, most just hate fueled things I've seen in a while. Like, 'DefendingAiArt' by actively threatening to do... what? Train a model on it just to be a awful person? Just, what does this person gain from this?