07mk
5d ago
I don't think LLMs can generate meaningful human-like feedback of what it feels like to use the software. They just don't see the UI in the way that humans do.
I don't see why LLMs would need to "see" the UI in a way similar to humans in order to generate meaningfully useful feedback for improving the UI (as well as any other element of the software) as judged by humans. It's not like the LLM would need to reason out "this UI element here gets in the way of this process due to that issue, etc." or "in my experience of trying to use this software in my workflow, this UI element could be improved by moving it here," or whatever. It'd be doing naked dumb pattern matching, of predicting words based on the prompt (which would include the sequence of 1s and 0s that make up the software, as well as instructions to produce text that a helpful human tester would provide, or the like) and its weights. There's no proof that this would work, but I also see no reason why simply scaling up current techniques and/or making them faster wouldn't allow LLMs to generate feedback like this which is just as useful as human user feedback.
07mk
5d ago
Yes, Adobe Premier is a few million lines of code, and LLMs can create millions of lines of code within weeks. However, Adobe premier wasn't one-shotted by a person, and an agent can't one-shot it either. The only way to build an excellent enterprise tool is to build a shitty enterprise tool, get feedback, and improve it with time. In startup speak, this private feedback is referred to as 'moat'. LLMs make this loop faster, but you can't skip it.
The value in the text/images/media/any content that form the feedback comes from how modifying the software in a way guided by the feedback improves the software as judged by the people who gave the feedback (and people like them), not in the fact that content was generated by humans using the software and expressing their opinions. Generating the feedback that way through actual humans who used the software is a great way to ensure that that the feedback is valuable in this way, but I don't see why a sufficiently advanced LLM (or LLM-based tool) couldn't generate that feedback with just as much value (i.e. modifying the software in a way guided by that LLM-generated feedback improves the software as judged by the people who would have given the feedback, i.e. target audience), just by predicting the next word. And then modify the software through iterations until the feedback crosses some threshold of asking for small enough changes or something. I don't think this would be considered a "one-shot," but it certainly seems like it would require almost as little investment in human effort. It's just that the LLM-based tools don't seem sufficiently advanced (or perhaps they're not sufficiently fast?).
07mk
5d ago
Lol. What were you doing on the platform?
Evidently, trying to produce videos that OpenAI disapproved of. More precisely, IIRC, I was inspired by some videos I saw on Sora where, apparently through some clever prompting and/or iterations, the user had managed to generate and share a video of a woman doing yoga, shown from suggestive angles. I was experimenting with doing the same when I got the ban email. Grok Imagine has lines as well, as I alluded to before, but it places its lines very very far from where Sora did.
07mk
5d ago
I wouldn't quite put it to the level of Shakespeare words. Fortnight wasn't a common everyday word when I was a kid in the 90s, but most high schoolers would have been familiar with the word and what it meant, if memory serves, likely due to its usage in history class when reading documents relating to Revolutionary War or the Civil War. Which were only about 150-250 years ago, not 400 like Shakespeare.
I do wonder if kids these days know that Fortnite is a play on that word and what that word means, or if they just think it's some nonsense made-up word.
07mk
5d ago
Wow, that phenomenon seems really reminiscent of how "sexual assault" became a catchall term to be used when the speaker wants to create the connotation of forcible rape in contexts where the reality is some sort of harassment of a sexual nature. As well as how "sex trafficking" became a catchall term to be used when the speaker wants to create the connotation of kidnapping women into sexual slavery in contexts where any level of prostitution took place. I've come to really dislike these overt attempts to engineer language for the purpose of hiding covert attempts to manipulate others into believing things that one finds useful for others to believe and personally always just call them "child porn" and "revenge porn."
07mk
5d ago
On the other hand, most of the effort with a commercially viable video generation product is in the product engineering, not the model itself. That's asking a lot lot of effort from OpenAI in an area they are not best equipped to beat seasoned product engineering teams at.
It seems evident by their actions that engineers at OpenAI lacked the ability or capacity to use GPT5 to cost-effectively write an Adobe Premiere competitor but with Sora-integration, with UI that's just as good, just as intuitive and user-friendly for longtime video editing professionals, just as stable and responsive, etc.
I wonder if/when AI companies will reach the point where they could just do that for any arbitrary existing software. At what point could one of these companies just instruct the AI to generate an Excel clone that has perfect backwards compatibility to MS Excel, but also has their AI integrated in, and consistently get out a viable software product as the result? What about a Windows clone that has perfect compatibility to all Windows-compatible software, but also has their AI integrated in? What about an Oblivion clone that has perfect compatibility to all existing save files, but also doesn't require major QOL overhaul and performance mods to make enjoyable and also has their AI integrated in?
07mk
5d ago
It appears to me that these issues could probably be solvable. Disallowing generative editing of user-uploaded images seems like a no-brainer.
Grok Imagine has indeed implemented something like this, where uploaded images of real humans become extremely difficult to edit without triggering a censor, and I think videos might be right out. Unfortunately, even this gimped censor is severely limiting, and a full-on prevention of generative editing or animation of user-uploaded images would take away like 90% of the use cases for image/video generative AI. Since so much of using gen AI to produce images and videos is about trial-and-error and iterations of manual edits -> AI generation building on it -> manual edits of AI generations -> AI generation building on it -> etc., including using multiple different non-interoperable AI tools (e.g. generate original image in Midjourney, edit it locally using Krita and Stable Diffusion, then upload it to Grok AI to animate), lack of ability to take arbitrary image input would leave it as only the origin point for the workflow, which doesn't amount to much, or just simple time-waster slot machine generations.
Personally, I do not like the notion that it's possible to arrange pixels in an illegal way that doesn't involve some other independently illegal action as a causal factor and hope that attempts to make it so fail horribly. Unfortunately, I'm not hopeful, as it seems to me that support for free speech and free thought isn't very high right now in the USA. This is also why I'm still holding out hope that video gen on local hardware will become "good enough" that private servers owned by companies like OpenAI, xAI, Google, etc. don't become effective gates for this sort of creative endeavor for the layman (or at least lay enthusiast).
07mk
6d ago
I tried using Sora for about a month at the end of last year, but I had to stop due to getting banned. Grok Imagine wouldn't ban me, so I've been using that instead. My wild guess is that a social media platform based entirely around AI generated videos like Sora can only exist in a sustainable way if it's explicitly for erotic/pornographic material - there's simply not enough demand for creating or viewing AI generated videos that aren't in that category to get enough users and views to pay for the generations.
I've mixed feelings, since Sora was clearly much better and more flexible than Grok imagine, and so I would've loved to see that develop further, but, at the same time, the lower censorship in Grok and XAI's general attitude towards censorship versus OpenAI's makes me think improvements in Grok is more likely to bear fruit. Of course, without Sora around, XAI has less reason to improve Grok... And Grok Imagine is also still censored, which isn't great, but it's the least worst, at least. In the long run, I'd hope that local video generation will be "good enough," but that'll probably require a world where dual 5090s with 64gb VRAM is considered a quaint little living room computer for sending emails and running old games at a tolerable 25fps, which I'm guessing is within 2 decades.
Reading your response before reading the quoted part (though I caught the capitalized names in the quote and registered them mentally), I had assumed the phrase was meant to invoke someone laundering the current Overton Window as a way to crystalize it where it is, rather than someone expanding or shifting the current Overton Window. Since we don't have an equivalent job for making windows bigger or moving them around in a wall (though hopefully nanotech will be so cheap in the future that using it for some silly feature like sliding the windows in your home to any arbitrary configuration at any time will be taken for granted), I can't think of a term that I'd find more appropriate.
More options
Context Copy link