Here's What Gemini Thinks About the Whole Creative AI Argument

[ghettosynth]

5 paragraphs justifying AI making sh*t up and blaming users for it. This is exactly why lots of people aren't impressed with AI.

The expectation that a probabilistic model that predicts the next token will not incorrectly predict the next token is a very big part of the problem. Hallucinations are an intrinsic part of how language models work. They have no internal state that reflects ground truth. Their state, such as it is, is reconstructed on each turn with the existing text that is in the chat.

So, unfortunately yes, you SHOULD expect it as as a user and if you don't expect it then you are using language models with unreasonable expectations. Also, yes, skill in how you interact with models will impact the "correctness" of the responses. If you want to call that "blaming the users", then knock yourself out.

Like regression models, next token prediction models are simply a calculation. Inappropriately applying a calculation to a user problem is a user error. If you're not impressed with them, then don't use them, and your problems will be solved.

[ghettosynth]

AI cant and doesnt ‘Think’ and I concur with the AI synopsis of Zerocrossing’s assessment of it.

I agree. This is really quite instructive. AI doesn't know zerocrossing's internal state, however, it can read, at least some of, zerocrossing's words.

I asked ChatGPT a similar question, the report isn't interesting, although it's similar and also looked up and read various posts here. Then I asked it, specifically, how such queries can result in hallucinations. It gave a detailed synopsis, in particular, highlighting how these specific kinds of queries are very likely to generate hallucinations. I then asked it how a responsible user could mitigate these concerns. It gave some good advice, I won't repeat it here, ask it yourself. I then used one of the more advanced levels to discuss its own results.

The first response was the weak link. It appears to have mixed real KVR material with wrong attribution.

The clearest error is the line about ChatGPT not knowing anything and not being able to judge factual accuracy. That text exists on KVR, but it was posted by starthief, not zerocrossing. The line about the debate “never really resonated” and the remarks about future “space for human expression” also appear on KVR, but they belong to Lunarkittn, the thread starter in the 2026 discussion, not to zerocrossing.

What zerocrossing actually said in that 2026 thread is noticeably different. He argued that AI is closer to human creative recombination than critics admit, said the main difference is intent, rejected the claim that AI can only arrange things from the past, and even asked whether copying feeling might count as expressing feeling. He also posted positively curious things about AI voice tools in 2023, and later criticized technical artifacts in an AI voice product in 2025. So the first answer’s bottom-line characterization of him as mainly skeptical was, at minimum, incomplete and in key places wrong.

The second response, about how a query like that can hallucinate, was broadly accurate. The described failure modes fit the standard picture in the literature: LLMs can generate fluent but unsupported claims, and person-specific summaries are especially vulnerable to attribution errors, invented support details, and narrative smoothing.

The third response, about how a responsible user mitigates this, was also broadly accurate. Using AI output as a lead rather than a source of record, checking primary sources, and verifying precise details are all aligned with mainstream AI risk-management guidance.

So the clean summary is this: the meta-level answers were mostly sound, but the original factual answer about zerocrossing was not reliable. It likely hallucinated by pulling together real KVR discussion fragments from multiple users and laundering them into a tidy profile of one person. A better corrected summary would be: zerocrossing seems mixed rather than simply anti-AI. He has raised criticisms of particular AI tools and their artifacts, but he has also argued that AI creativity is not as fundamentally different from human recombination as some critics claim, with intent being the key distinction.

Note, that this may still be incorrect. However, some of this information is reliable and can be reasonably trusted at a high level, and some of it cannot without verification. Which is which is the user's responsibility to determine. What you should not be naïve about is this. Despite that there is widespread inappropriate use of AI, especially in the work place, this level of accuracy is often sufficient. The comparison is how effective would a human be at this, not whether or not it's correct.

I then asked it so summarize the KVR sentiment on AI without invoking any users's names.

Overall, KVR’s sentiment toward AI in music looks mixed, active, and somewhat wary rather than uniformly hostile or enthusiastic. The site now has a dedicated “Machine Learning and AI for Music Creation” forum whose description explicitly frames AI as something that can expand creativity while keeping the human in the creative workflow, and that forum is busy enough to host both highly skeptical thread titles and openly enthusiastic ones. On March 9, 2026, its front page showed threads such as “AI? LMAO...We’re Not There Yet,” “Can AI (Artificial Intelligence) even compete?,” “If AI replaces musicians, does the entire plugin industry die with them?,” and “SUNO is killer!,” with some drawing hundreds of replies and one surpassing 1,000.

The clearest pattern is pragmatic interest in AI as an assistive tool, alongside strong unease about AI as a replacement for musicianship. In practical workflow discussions, people treat ChatGPT and similar tools as useful for things like patch design, SysEx generation, and ideation, often as a starting point rather than a finished result. In broader debates, the recurring concerns are originality, authorship, ownership, harm to working musicians, and whether full automation would damage the plugin ecosystem itself by shrinking the population of human creators.

There is also a noticeable split between assistive AI and fully generative AI. Threads about AI-assisted mixing, patching, and vocal tools tend to sound more accepting, while threads about fully AI-generated songs and long-term industry consequences pull in more ethical and existential anxiety. One discussion about the future of AI in music captures that balance neatly: AI is described as potentially very helpful for mixing and production tasks, but current music generators are also criticized for audible artifacts and limited quality, with the longer-term fear being a flood of machine-made content.

So the compact read is: KVR is curious, divided, and engaged. The community seems fairly open to AI that helps musicians do their work, but much more skeptical when AI is framed as replacing creative labor, muddying authorship, or flooding music platforms with automated output.

So ask yourself, is that pretty close? Then ask yourself this, could this data be useful to someone at some point? Then ask yourself, if it's slightly wrong, would it matter? Then, if you had to write this summary, how accurate would your summary likely be, and how long would it take you to write it?

[zerocrossing]

5 paragraphs justifying AI making sh*t up and blaming users for it. This is exactly why lots of people aren't impressed with AI.

I think what we’re seeing is people who are either not very creatively inclined, or are insecure about their ability to make art, use AI as a way to say, “See? What you do wasn’t all that impressive, all along. A machine can do it just as well!”

Again, follow the money. There’s a ton of it flowing into AI investments, and nothing flowing out. Nvidia doesn’t have a return policy, so it’s salad days for them. On the other side, I see a lot of hand waving when people ask very simple questions, like who is going to pay for AI if it replaces all the office and software engineering jobs. Yet, at the same time, I’m seeing ads for software that helps people fix the problems that AI agents make.

Matt and Trey Parker nailed it when they did the episode where the kids figure out how to get ChatGPT to write their papers. Mr. Garrison is overwhelmed by the volume of writing, so he uses ChatGPT to grade and comment on the papers. Don’t worry, though. No one will need to know how to read and write, because Sam Altman says that GI is on the horizon and it’s going to take care of all of us. We just have to keep giving him money.

[ghettosynth]

5 paragraphs justifying AI making sh*t up and blaming users for it. This is exactly why lots of people aren't impressed with AI.

I think what we’re seeing is people who are either not very creatively inclined, or are insecure about their ability to make art, use AI as a way to say, “See? What you do wasn’t all that impressive, all along. A machine can do it just as well!”

What does that have to do with my response?

On the other side, I see a lot of hand waving when people ask very simple questions, like who is going to pay for AI if it replaces all the office and software engineering jobs. Yet, at the same time, I’m seeing ads for software that helps people fix the problems that AI agents make.

What you are seeing is how any new technology is integrated, often incorrectly at first. AI will not replace all of the office and software engineering jobs, AND, you will need to manage problematic agentic output. That people are selling software to do this is just a statement about people identifying a problem and trying to sell a solution. It says nothing intrinsically about their success in doing so, nor does it necessarily identify the actual problem, per se.

[BONES]

5 paragraphs justifying AI making sh*t up and blaming users for it. This is exactly why lots of people aren't impressed with AI.

What I got from it is that it's people who expect AI to behave like a human who seem to struggle with it, while people who see it for what it is get a lot more out of it. That's kind of paradoxical, in that it's the first group who seem desperate to show that AI is no substitute for a human, something that the people they are trying to convince already understand.

[jancivil]

What does that have to do with my response?

Why should it have to? It's not even really the same subject, he's on about music-"generating" AI while you're explaining the nature of LLM. You seem quite knowledgable on that front, but the topic is 'the whole creative AI argument', not 'ghettosynth explains why you're wrong about LLM getting shit wrong'.

[ghettosynth]

5 paragraphs justifying AI making sh*t up and blaming users for it. This is exactly why lots of people aren't impressed with AI.

What I got from it is that it's people who expect AI to behave like a human who seem to struggle with it, while people who see it for what it is get a lot more out of it. That's kind of paradoxical, in that it's the first group who seem desperate to show that AI is no substitute for a human, something that the people they are trying to convince already understand.

Exactly correct.

[BONES]

If you fire off enough rounds, you're bound to score a hit eventually.

[BONES]

I think what we’re seeing is people who are either not very creatively inclined, or are insecure about their ability to make art, use AI as a way to say, “See? What you do wasn’t all that impressive, all along. A machine can do it just as well!”

Conversely, I see people who are dismissive of AI in some (or many) ways as insecure about their creative ability, refusing to acknowledge that a mere machine might be just as capable in that department as they are (and will be more capable than them in the near future). Some are so afraid of AI that they aren't even willing to have a go and find out for themselves. Meanwhile, the other group have experienced both sides of the argument and are much better placed to make an objective assessment.

Again, follow the money. There’s a ton of it flowing into AI investments, and nothing flowing out.

That's called R&D and/or Product Development. The pay-off comes later. Car companies spend a billion dollars developing a new model before they sell even one. It's how the world works.

On the other side, I see a lot of hand waving when people ask very simple questions, like who is going to pay for AI if it replaces all the office and software engineering jobs.

That's the dumbest question I can think of. Who used to pay the wages of all those people AI replaces? Their margins will increase by an order of magnitude which will allow prices of many goods and services to be a fraction of what they are now, so we'll still be able to afford things when we're all on Universal Income. If you're really worried, though, now might be a good time to start growing your own food. Two or three chickens per person will provide enough eggs to meet all your protein needs, maybe with some left over to barter with your neighbours.

[zerocrossing]

What does that have to do with my response?

Why should it have to? It's not even really the same subject, he's on about music-"generating" AI while you're explaining the nature of LLM. You seem quite knowledgable on that front, but the topic is 'the whole creative AI argument', not 'ghettosynth explains why you're wrong about LLM getting shit wrong'.

The funny thing about this is, I've had ghettosynth and BONES muted for years because their posts are generally rubbish, and only see them when they're quoted, so all I can think of is, "you're so vein, I bet you think this post is about you..."