At least we don't have to worry about AI taking over any time soon
Sora is here
921–930 of 1001 posts
Re: Sora is here
#922Every day that passes I grow fonder of Google's decision to delay or otherwise keep a lot of this under the wraps. The other day I was scrolling down on YouTube shorts and a couple videos invoked an uncanny valley response from me (I think it was a clip of an unrealistically large snake covering some hut) which was somehow fascinating and strange and captivating, and then scrolling down a few more, again I saw someth…
This is all inevitable. At worst it's pulling the issues forward by a few months or years, and I don't think anyone will meaningfully address the problem until it's staring us in the face.
I believe the internet needs a distributed trust and reputation layer. I haven't fully thought through all the details, but:
- Some way to subscribe to fact checking providers of your choice.
- Some way to tie individuals' reputation to the things they post.
- Overlay those trust and reputation layers.
I want to see a score for every webpage, and be able to drill into what factored into that score, and any additional context people have provided (e.x. Community Notes).
There's a huge bootstrapping and incentive problem though. I think all the big players would need to work together to build this. Social media, legacy media companies, browsers, etc.
This also presupposes people actually care about the truth, which unfortunately doesn't always seem like the case.
Re: Sora is here
#923Every day that passes I grow fonder of Google's decision to delay or otherwise keep a lot of this under the wraps. The other day I was scrolling down on YouTube shorts and a couple videos invoked an uncanny valley response from me (I think it was a clip of an unrealistically large snake covering some hut) which was somehow fascinating and strange and captivating, and then scrolling down a few more, again I saw someth…
I saw my first AI video that completely fooled commenters: https://imgur.com/a/cbjVKMU This was not marked as AI-generated and commenters were in awe at this fuzzy train, missing the "AIGC" signs. I'm quite nervous for the future.
This is definitely something the Japanese would do, but it is not a real train unless a thousand salarymen are crammed into it.
Re: Sora is here
#924How long until they fix the sign up issue? What an embarrassment. Why release something if you know it can't work properly? And why do we need to sign up when we already have an account with ChatGPT?
It was cool when they announced it but the novelty of generating a piece of AI video clipart is quickly fading, especially when it takes months or years to just get a demo in users' hands.
Re: Sora is here
#925Every day that passes I grow fonder of Google's decision to delay or otherwise keep a lot of this under the wraps. The other day I was scrolling down on YouTube shorts and a couple videos invoked an uncanny valley response from me (I think it was a clip of an unrealistically large snake covering some hut) which was somehow fascinating and strange and captivating, and then scrolling down a few more, again I saw someth…
It saddens me. Innovations in AI 'art' generation (music, audio, photo) have been a net negative to society and are already actively harming the Internet and our media sphere. Like I said in another comment, LLMs are cool and useful, but who in the hell asked for AI art? It's good enough to fool people and break the fragile trust relationship we had with online content, but is also extremely shit and carries no meani…
There was a recent discussion in another HN thread that I think summed it up well. Good art rewards a careful viewer; the more you look at and think about good art, the more you get out of it. AI art does the opposite and punishes thoughtful consumers. There's no logical underpinning to the various details, it's just stuff mashed together in a superficially nice looking way.
Re: Sora is here
#926Earlier quoted context omitted.
I agree. Apart from the text appearing backwards it all looked pretty real to me.
Yeah, that's a weird one. I doubt the video was generated that way. I assume someone flipped the video for "artistic" purposes.
Reversing text is a known loophole to getting around copyright guardrails in image-generation models.
Re: Sora is here
#927Earlier quoted context omitted.
Yes, the idea works and was explored with dreambooth/textual inversion for image diffusion models. https://dreambooth.github.io/ https://textual-inversion.github.io/
Both of those are of course out of date and require significant training instead of just feeding it a single image. InstantID ( https://replicate.com/zsxkib/instant-id ) fixes that issue.
Dreambooth style training is in no way out of date.
If you just want a face, InstandID/Pulid work - but it’s not going to be very varied. Doing actual training means you can get any perspective, lighting, style, expression, etc - and have the whole body be accurate.
Re: Sora is here
#928Every day that passes I grow fonder of Google's decision to delay or otherwise keep a lot of this under the wraps. The other day I was scrolling down on YouTube shorts and a couple videos invoked an uncanny valley response from me (I think it was a clip of an unrealistically large snake covering some hut) which was somehow fascinating and strange and captivating, and then scrolling down a few more, again I saw someth…
Shorts are designed to trade your valuable attention for trite, low-effort content. Most decent shorts are just clips of longer-form content. Do yourself a favor and avoid that kind of content, opting instead for long-form consumption. The discovery patterns are different, but you're less inclined to encounter fake content if you develop a trust network of good channels.
This is also my strategy. AI content makes me focus even harder on the source of the content instead of the apparent quality, because the current set of GenAI techniques are best at imitating surface-level quality features.
Re: Sora is here
#929Earlier quoted context omitted.
Yeah, that's a weird one. I doubt the video was generated that way. I assume someone flipped the video for "artistic" purposes.
Reversing text is a known loophole to getting around copyright guardrails in image-generation models.
How does that work? Would you prompt the model to write "hello Kitty but in reverse" on the train so the resulting image isn't flagged?
Re: Sora is here
#930Many people say:
> these things will get bigger and better much faster than we can learn to discern
I would like to ask “Why?”
Clearly, these models are just one case of “NN can learn to map anything from one domain to another” and with enough training/overfitting they can approximate reality to a high degree.
But, why would it get better to any significant extent?
Because we can collect an infinite amount of video? Because we can train models to the point where they become generative video compression algorithms that have seen it all?