MelomaN #‎ Sharp 🅥

CHELSEA ARE THE BEST! 💙💙💙

In these dark and crazy times there are few reasons for joy, but miracles inspire. And I am more than ever glad that Chelsea is my favorite club.

1 month ago | [YT] | 0

MelomaN #‎ Sharp 🅥

Solving all the problems with the dominance of AI content on the Internet would actually be very simple, if there was a desire.

I'm not talking about banning AI-generated content or filtering it. Those are crude and ineffective ways to solve the problem. This issue needs to be addressed at the level of how AI is approached, how it is constructed, and how access to it is provided.


And no, I'm not talking about the meaningless mantras that "AI is incompatible with creativity." I've always considered this viewpoint to be nonsense for one simple reason: in almost any creative process, there are routine, repetitive actions not directly related to the creative spark that SHOULD be automated. These are tasks where you don't come up with something that brings you inner joy, but rather do something that just needs to be done to achieve the final result. When you design and build something in a game like Minecraft, envisioning its layout, you'll encounter not only the need to think through its elements and internal planning, pondering which Minecraft blocks can best represent modern elements, but also the need to spend hours placing thousands of blocks for necessary walls. (This problem is easily solved with tools like mcedit.) A producer has to deal with purely technical tasks like cleaning tracks of noise (which AI tools can do). An animator doesn't need to manually redraw a face if they just want to copy it (face copying technology is available, for example, in Daz3D). These tasks are exactly what should be delegated to AI—and this wouldn't be "harming creativity," but rather freeing it from routine parts, allowing more creation.

However, arguing against the view that neural networks for the internet are more of a cancer than something useful is becoming increasingly difficult, even for me, who usually argues against this. I must admit that I am biased in this matter; by education, I am a programmer (my bachelor's specialization is called "applied mathematics and computer science"). As a result, I have always had a positive attitude towards machine learning and artificial intelligence, and spoiler alert, I still do. But not in its current form.

A few seemingly obvious thoughts about generative AI:

1. There should be a high barrier to entry for using AI.
2. It should be freed from meaningless and ruthless censorship, which affects its quality and ultimately increases the templated nature of the content it generates.
3. If we are talking about narrow AI, its main goal should be to be maximally customizable as an "extension" of the strengths of the specific person using it, not as a machine that clumsily imitates a human and tries to "replace" them, failing in the process.

If these three principles were widely applied, there would be no talk of theories about the "dead internet." And incidentally, there would be fewer issues with job displacement, but I don't even want to get into that seriously.

Now, let me explain in more detail.

When I talk about a high barrier to entry, I'm not necessarily saying "raise prices to several thousand dollars for a subscription"; that's just one option and not the best one. What I mean is that generative AI should not work on a "one-click and get a result" system.

Surprisingly, China is the best example of how to approach this problem. Chinese developers, from what I observe, are turning generative AI into a development environment. This is exactly the role that narrow generative AI should play: it should be a widely available toolkit that enhances the qualities you already have as a creator and helps realize ideas that you couldn't before or that would take too long to implement.

In this context, a high barrier to entry is not an artificial complication, but a natural barrier against turning AI into a factory for generic trash. Just as any musical instrument requires at least basic knowledge to play something meaningful, generative AI should require an understanding—even if superficial—of what you are doing, why, and how it should work.

The second aspect is censorship. This is not about "letting AI do whatever it wants"; it's a logical continuation of my first point. If generative AI is an extension of the human, then "sanitizing" it through filters and bans is an attempt to suppress not the AI, but the author. The problem here is not that AI "can't generate something edgy." The problem is that under active censorship, the neural network learns to avoid unconventional thoughts altogether, making the content more homogeneous, templated, and low-quality. It starts to suppress the author's voice, replacing it with a bland, institutional tone. Such a system becomes not an assistant, but an agent of institutionalized blandness. Moreover, filtering affects not only "dangerous" content—it directly influences the artistic, stylistic, and emotional parameters of the output. Edges are smoothed. Structures are simplified. The result is templated content, not due to a lack of power or design of the model, but out of fear that it might say something wrong.

The third and perhaps most important point is that narrow AI should be an extension of the author themselves. In practice, this means that AI should be trained—or at least calibrated—on your personal works, styles, motifs, and developments. This solves a lot of problems. First, it addresses the issue of "uniqueness"—content generated by such AI is inherently part of the author's language. Second, it greatly increases relevance: if you train AI on your prose, drafts, and visual works, it starts helping you in your style, not offering something vaguely "similar." Third, it elevates AI from imitation to synergy because it stops being a replacement and becomes part of your creative process.

This is what narrow generative AI SHOULD be in a perfect world. This, by the way, solves not only the problem with creativity but also with academic plagiarism.

The "one-click and get a result" system only seems tempting. In reality, the result is one: people appeal to the theory of the "dead internet" and massively hate AI due to the proliferation of homogeneous content because we pass it off as creativity. When the result of a click begins to be perceived as "content," it destroys not only originality but also authorial involvement itself. The only situation where this approach might work is in the future: when we entrust the task to AGI. AGI will actually be able to understand the task, build a hypothesis, and solve the problem comprehensively, including having modules that perform the functions of the limbic system (without them, it won't be a full-fledged AGI due to the lack of such necessary human traits as tastes and a sense of beauty—for this, it needs its own will). But the catch is that AGI will almost certainly not deal with such trivialities. AGI will be needed for tasks that humans CANNOT solve. And it will not be a tool, but an intellectual partner who will have their own views and, possibly, priorities.

Only such an approach to generative AI, in my opinion, can be healthy. THANKFULLY, it is already being successfully implemented to some extent, for example, in professional animation studios in Japan. But unfortunately, there is a feeling that Pandora's Box has already been opened, and its contents cannot be put back.

Well, we'll see what happens.

3 months ago | [YT] | 5

MelomaN #‎ Sharp 🅥

An easy test to see if you're a boomer. (Yes, if you answer correctly).

"I always feel like..."

4 months ago | [YT] | 0

MelomaN #‎ Sharp 🅥

For those who subscribed to this channel because of WiS-edits, unsubscribe and re-subscribe there (in the future WiS-edits can be expected only there): youtube.com/@MelomaN-ReVerse/shorts
I just finally figured out how YouTube's recommendation system works, and if that's really the case, posting them here any further is simply destructive.

4 months ago (edited) | [YT] | 0

MelomaN #‎ Sharp 🅥

First you have to wait 6 hours for someone to see your new shorts, and then you post another shorts and get 1000 views in 15 minutes. I give up trying to understand YouTube's algorithms.

4 months ago | [YT] | 3

MelomaN #‎ Sharp 🅥

Dude literally made his homies billions richer because of the market volatility he HIMSELF CREATED! The chart shows the volume of purchases: when people were tearing their hair out from the fall, his insiders were buying.

He's much better at looting money than at peacekeeping.

4 months ago | [YT] | 2

MelomaN #‎ Sharp 🅥

I have never seen a person who was more right before @HamburgerCybercotleta

4 months ago (edited) | [YT] | 5

MelomaN #‎ Sharp 🅥

ara-ara

4 months ago | [YT] | 4

MelomaN #‎ Sharp 🅥

Command & Conquer: Blue Alert 2

4 months ago | [YT] | 1