It's not "neutral machine-based enforcement", it's a general purpose panopticon that tracks and logs everything, ready to be queried later by arbitrary people for arbitrary reasons.
18,061 karma · joined June 2, 2013
You really think if I asked 200,000 people in my city, only one would have heard of Meta RayBans? when there are billboards, magazine ads, and window stickers in every glasses store? Couldn't tell you the real number but it's probably closer to 1 in 5.
Not uncommon, in comments like these, for there to be so many 9s that the population of Earth is exceeded.
Graphing Calculators are computers in every sense, it's hard to find one you can't at least program in a high level language like Basic or Python, and many accept assembly programs - even Texas Instruments models did, before their signing keys were brute forced.
Games consoles too almost invariably run third party code, although they tend not to be end user programmable - no different than an iPhone in that regard, really.
"Feature phones" - pre-smartphone phones with cameras and internet etc - usually had the ability to run Java Midlets without manufacturer blessing, even though you couldn't touch the OS. You could SSH into servers from those!
I think a big problem is that the distinction between "general purpose computer" and "end user programmable general purpose computer" is functionally meaningless when you can't write programs, which is most people. Instead they understand it in a functional context - what is this device for? How does it compare to how such devices usually work? The general purpose CPU inside is an implementation detail, and just because you think of an iPhone as a computer doesn't mean everyone does. They think of it as a "phone".
But, again, you cannot consider flawed human perception to be the ground truth! Photonically, the red eye is the ground truth. If you were looking through the viewfinder, you should also have seen the red eye effect, if only for an instant. If you didn't, it's only because your eyes were too slow. Or perhaps you weren't looking through the viewfinder at all and therefore not in a position to see the retroreflection, in which case you might as well have been facing the other way for all that counts.
Or consider taking a photo from a moving vehicle. You can't resolve your blind spots with saccades because the scene moves. Your subjective impression is probably that only the thing you're tracking is crisp, and everything else is a vague, blurry mess. Is that what the camera should capture, irrespective of shutter speed?
I think it's clear that culturally we expect photographs to be an indexical trace of genuine photons, not a replication of subjective experience. That is the value they provide over an "artist's impression".
There is no dividing line between syntax and semantics - semantics is just syntax scaled. Godel proved it, LLMs exploit it.
"what the human originally taking the photo saw with their naked eye" is a tempting standard, but a misleading one - there might not be a human eye near the camera at all, and at any rate there certainly isn't one co-located with the focal point. And shall we have cameras that render only a few degrees of high resolution color, and have a large-ish blind spot? It's what the human eye sees.
The problem is philosophical - people are upset by certain types of enhancement error, but not others, and they can't even articulate the difference to themselves, let alone to an AI.
In the long run... well, general purpose computing is under threat anyway. Platforms with locked bootloaders and mandatory digital signatures outnumber those that don't. Even on "PCs", the openness is merely a cultural norm observed for a particular market segment[0] by Apple and Microsoft - for they are the only ones with true root keys. Even a brand new motherboard comes with Microsoft signing keys pre-flashed, and you are permitted to enroll your own only by grace. The technical infrastructure to flip a switch and lock it all down is now in place, ready to go at the stroke of a pen. Will you be able to get Llama.cpp from "the app store"? I hazard not. If you think this sounds hyperbolic, look at what is happening to Android.
[0] They don't even segment it the same! Apple segments by "does it have a keyboard", and Microsoft segments it by "does it have a native x86 processor". Apple runs different software on the same basic hardware, Microsoft runs the same basic software on different hardware, but both have created artificially restricted second class computing categories.
No, I will not chill. The war on general purpose computing is gonna get real hot real soon.
- watermark-free generation
- the stripping of watermarking from the output of SAAS models
Any discussion of watermarking is dead in the water in a world where we are permitted to have these things. I fear for the future.
Signatures are extra data, added out-of-band to the existing data. Out of band data, by definition, is easily detected and stripped, so the utility of a signature is that authoring one requires secret knowledge. Philosophically, the presence of a signature is a kind of authentication, a desirable thing that is hard to grant and easy to revoke (the smallest change to the data renders it invalid).
Now, watermarks: if you flip it round and say you want to glue on a piece of undesirable data - something that represents disauthentication, like a cursed black spot of written-by-LLM - then you want it to resist removal efforts. And now right away you have a hard problem because your sticky data must be in band, or else it is trivially stripped. Not only that, in fact, it has to look enough like real signal that it isn't easily filtered. And on top of that, you can't distort the real signal too much, or people will complain. So you're cornered into doing a kind of steganography - hiding small amounts of information in the entropy, biasing the signal in perceptually plausible ways that are detectable to those in the know. Cartographers add fake streets ("trap streets") to catch plaigiarists - for LLMs, watermarking might take the form of subtly odd word choices.
Looking forward to nitter-style reddit proxies springing up so we can actually read the information freely contributed by decades of unpaid volunteers.
Because they instantly and reliably(tm) give you the information you want in the format you want it in, they are going to replace the web search as the dominant mode through which information is disseminated - if, indeed, this hasn't already happened. As such, LLMs are effectively a form of publishing for their training data.
Which means a vicious fight for control over what goes into LLMs, and strong incentives to widely disseminate useful LLMs that encode your structural biases, instead of some adversary's. We already talk about Chinese models that won't talk about Tiananmen Square, but that's eye-rollingly sophomoric compared to what the stakes are now. Imagine a model that subtly discourages entrepreneurship, because its creator doesn't want upstart competitors. All questions of media bias are multiplied incalculably when LLMs are folded into all of society.
I think we need to take LLMs seriously as sovereign projects that are as critical to the democratic experiment as a free press. Funding needs to be nationalized; training needs to be conducted in the open, on open datasets; fine tuning needs to follow democratic principles. Only then will they be fit for purpose for their inevitable structural destiny: arbiters of human opinion.
Once upon a time, I encountered a sort of interactive museum book called "Earthsearch". One of its "exhibits" was a color wheel, made of two spinnable pieces of tinted acetate, that purported to be able to generate any skin tone. I suppose this is three degrees of freedom, since for any position of the two wheels you could access a range of colors around the edge - but the printed page underneath was simply a grayscale gradient, so one degree of freedom was simply luminance.
The book was released in 1994, and was very confident. It even went as far as to encourage the reader to see a doctor, should their skin tone not be achievable! What color model were they using?