A 2024 Discussion Whether To Convert The Linux Kernel From C To Modern C++

[ssokolow]

I didn't intend to hint that (I guess you got that). The underlying problem is general ignorance and lazyness of people.

• The lazy student that doesn't invest in real work and uses A.I. to write his paper
• The lazy prof that doesn't want to read the paper and just uses A.I. to disqualify it
• The lazy insurance clerk that answers your complaint with "the computer did it"
• The artist that deliberately produces crap because he can get away with it. (and probably teases himself with the reactions of the "experts")

The problem is that you're not going to stop humans being humans. We're hard-wired to try to economize nutrients to our brain. That's one reason we have so many wetware exploits and why we stereotype and so on.

[Anux]

The problem is that the same argument could be levied against digital art because the people who make it don't understand how to mix paints or handle a chisel or what have you and probably was levied against them initially.

That's not an argument and also totally different from generating random/AI images and then selecting 1 out of 1000 till you find something beautiful. This leads to not involving any art making process, because art starts in your head and everything else is just the tools and restrictions you use to bring the art to paper.

AI has no means of critical thinking, self-reflection or even understanding anything (human language, society).

[lowflyer]

The problem is that you're not going to stop humans being humans. We're hard-wired to try to economize nutrients to our brain. That's one reason we have so many wetware exploits and why we stereotype and so on.

In software there is a distinct benefit of lazyness: It can be a driving force behind optimization. Well humans will always remain humans and errors will be made. I will always accept that. However we can (and should) call them out for ignorance - or "willful blindness".

Everything should be made as simple as possible, but not simpler.

[Old Grouch]

That's not an argument and also totally different from generating random/AI images and then selecting 1 out of 1000 till you find something beautiful. This leads to not involving any art making process, because art starts in your head and everything else is just the tools and restrictions you use to bring the art to paper.

AI has no means of critical thinking, self-reflection or even understanding anything (human language, society).

How do you know anyone is capable of "critical thinking, self-reflection or even understanding anything (human language, society)"? Think of the concept of a philosophical zombie, which is not a joke.

[Anux]

How do you know anyone is capable of "critical thinking, self-reflection or even understanding anything (human language, society)"?

Easy, just talk to them about their life.

[Old Grouch]

Easy, just talk to them about their life.

Your understanding of the concept of a 'philosophical zombie' differs from mine.

As a means of distinguishing an LLM pretending to be human from a human, your suggestion has merit, mainly because most LLM's context windows are too small.

[Anux]

Your understanding of the concept of a 'philosophical zombie' differs from mine.

I don't have any apart from it being a philosophical theory. If I am talking to a real human I can easily verify if he understands what I say and that he is not a computer. And by talking about his life and decisions throughout I can easily notice if he has any critical thinking and self-reflection. And those are characteristics needed to produce any meaningful art.
Otherwise you're just creating beautiful things at best, like colorful flowers.

As a means of distinguishing an LLM pretending to be human from a human, your suggestion has merit, mainly because most LLM's context windows are too small.

It's not about context windows. It's about understanding what is talked about rather than making statistics based decisions. The latter will always throw together things that don't belong together simply because the statistic says it's probable.

[ssokolow]

That's not an argument and also totally different from generating random/AI images and then selecting 1 out of 1000 till you find something beautiful. This leads to not involving any art making process, because art starts in your head and everything else is just the tools and restrictions you use to bring the art to paper.

AI has no means of critical thinking, self-reflection or even understanding anything (human language, society).

As someone who dabbles with Stable Diffusion for my own private use (I use it as an "automated brainstorming helper" for getting me out of writing-hobby ruts in the same way that a linter is an "automated code reviewer". Seeing where the A.I.'s preconceptions differ from mine for the same prompt is useful for breaking out of ruts.), I can tell you that's not how anyone who actually has any skill with it uses it. It's more a cross between being a nature photographer (i.e. having the skill and patience to recognize good composition and potential for future tweaking) and things people who are unarguably are artists say, such as that their work was always in the wood/marble/whatever and they just exposed it, or the surprising amount of skill at painting that goes into taking a circuitous route through the vagaries of the paint, how it mixes, the canvas, and how the paint interacts with it in order to arrive at what you envisioned... modified by "happy little accidents" along the way.

For example, some of the things about Stable Diffusion that aren't just "push button until your receive art" include:

• Refining prompts (You'd be surprised how many preconceptions these models have and how many hidden connections between different keywords... and that's before you get into things like using reverse prompting tools to actually find keywords for what you want, using model inspection tools to discover if the model training internalized a keyword or phrase by tokenizing it in an unhelpful way, etc. At the high ends, it starts to feel more like debugging crashes in a C program.)
• Gradient prompts (Similar to how you can animate properties in Blender 3D, you can tell Stable Diffusion to "animate" the weights of keywords in the prompt over the number of refinement iterations you specified, because Stable Diffusion rendering is sort of like progressive JPEG rendering or how painters will block in a canvas, then paint large blobs, then iteratively introduce finer and finer detail, so you can do stuff like telling it to start out rendering person A, and then switch to aiming for person B before the identifying details go in.)
• Regional prompting (If you want something to be true in one part of an image but not another, such as having two different characters in the same scene, you need to write multiple prompts and specify what portions of the image each one applies to. Not all Stable Diffusion frontends support this.)
• ControlNet (eg. OpenPose, which is the Stable Diffusion equivalent to putting IK bones inside a model in Blender 3D... though it's not always reliable and the tooling for it is still kind of byzantine.)
• Inpainting (Once you've got an image that's almost right, you can mark regions and re-render them, using the existing content as a biasing weight. Expertise deciding how to tweak the prompt applies. Inpainting may involve using the same model or a custom inpainting model more specialized to the task.)
• etc. etc. etc.

Note that both initial images and inpainting can also be used to turn blobs you draw into refined images, but that requires enough of a sense of proportion to be able to size and position the blobs such that you don't get a "garbage in, garbage out" effect where Stable Diffusion will choose to stick to the shape you drew, even if that means producing a deformed monstrosity. (I mainly use it to erase unwanted stuff in the background and then ask SD to please extend the background to naturally fill in the obvious-unless-you-squint-hard-enough copy-paste/paintbrush erasure. (eg. to erase a duplicate of the character I prompted for, because its conception of the requested setting included people in the background and, without regional prompting, it's likely to replicate the character description across each person it needs to draw.)

Overall, I'd say that art-generating A.I. is barely out of the IMSAI Altair Build It Yourself Kit phrase right now, but it's very much not "push button, receive acclaim" either way. It just looks that way because of the combination of hype and "Unlike traditional art tools, where having vision is easy and making it look good is the hard part", A.I. leans more toward "It'll try to make even the dumbest prompts look good, but they all quickly reveal themselves to be formulaic and boring unless you have a vision and the skill to successfully guide the A.I. into implementing it."

(i.e. The skill isn't in making a polished-looking output, it's in making an output that matches your vision instead of just being a bland re-hashing of whatever traits occurred most often in the training set... you can see why it's so popular for I-dont-want-to-pay-for-my-stock-images top-of-article filler.)

In a sense, what we're seeing is just a really fancy version of how, in the 90s, people would upload low-effort output from Photoshop plugins for generating things like marble/cloud/etc. textures.

[ssokolow]

In software there is a distinct benefit of lazyness: It can be a driving force behind optimization. Well humans will always remain humans and errors will be made. I will always accept that. However we can (and should) call them out for ignorance - or "willful blindness".

Certainly, but we mustn't fall into the trap of thinking we can fix individual people. That never works. We need systemic protections. That's why businesses get regulated. That's why colleges have a system for appealing bad marks. etc.

[lowflyer]

Certainly, but we mustn't fall into the trap of thinking we can fix individual people. That never works. We need systemic protections. That's why businesses get regulated. That's why colleges have a system for appealing bad marks. etc.

That's where I disagree. This is not the trap. You're correct saying *WE* can't fix individual people. We cannot. But we don't have to. People can only be "fixed" by themselves. And they do so. People turning themselves around and better themselves is the only way how society gets "fixed".

Blindly calling for "systemic protections" is the trap. It is pointing at "the others". It is the enshrinement of the saying "If only all would do like this / be like this ...". Too much regulations makes businesses go bankrupt. Too many ways of appealing makes college degrees useless. Who protects the regulated from the regulators? It's a big mistake to think that the rule-making-people are not "broken" themselves. Introducing rules is the communist/socialist/marxist approach. There are enough examples on this planet that show how this never works.

There needs to be a balance. While a ground layer of moral rules is necessary - anything much beyond that is from the devil.