Like, the technology behind this is no secret at this point. If not China, it will be Russia, or Belarus, or Iran, or whatever. Somewhere with electricity, servers, and a disregard for US intellectual proprietary laws. Also, even in China, I'm really skeptical that they would put filters for Chinese users (unless you are trying to make of Xi Jipinin that it is). The most likely outcome is that America/Europe get safeguards and Chinese users don't, which would lead to a funny situation of Westerners using Chinese VPNs to access a more unrestricted version of the model.
And that's not even to mention open source, which is the future of this tech. Open source, local, censorship-free models that you will be able to run on your home.
Yeah that's one of the real takeaways from this. This will improve over time. People seem to get so put off by hype that they forget there can be things of real significance underneath it. You could make a long list of what's amazing and promising about this "implement a browser" task, despite all its shortcomings.
The obvious? Selling subscriptions to individuals, reaching higher-ups with bombastic headlines, reaching potential investors, perpetuating the bubble.
This reminds me of a US District Court judge's ruling about Tucker Carlson on Fox News: "Any reasonable viewer 'arrives with an appropriate amount of skepticism' about the statements".
Sure, if you're talking about rubes who just got off the bus in the big city, then perhaps there's a point buried somewhere deep in the pile you're pushing. But is that who you think the HN audience is? Admittedly, talking to you, I can't rule it out.
Amusingly, by (dictionary) definition you're wrong, because "grift" means "engage in petty or small-scale swindling" (Merriam-Webster), but you're alluding to a much larger-scale phenomenon.
But it's tinfoil-hattish to claim that projects like this and the PR about it is part of a "grift". You're squinting hard to be able to see what you want to see.
>But what gives me joy with art is that it's a communication from one person to another
Maybe if a person generated 50,000 songs, not even listening to them, you could have a point. Although, even in that case, regardless of the lack of an "artist's intention," there is the interpretation of what people will take from that thing. And that interpretation is often different from what the author originally had in mind. Hell, most people don't know the author of most movies, TV shows, and the like they watched. In other words, to me, it's more about what people take from that thing, as opposed to "Oh, what that sentient being was trying to communicate?"
And I do believe a sufficiently advanced AI model would be able to mimic or synthesize human knowledge/worries/dramas in such a profound way that, regardless of "intention to communicate," it would be able to create things that people would relate to and take deeper meaning from.
Also: the very dataset where that thing was trained wasn't trained on an alien dataset, with an alien culture and the like, all originating from poems written by real people, movies by real people, etc., etc. The model learned from human culture; therefore, whatever it produces is a reflection of that culture, which people could and most likely will relate to, and, hell, they are already doing that.
But even taking the argument at face value, "Oh, human creation," someone might have used AI, but they were still involved in all parts of the creation process, like writing the lyrics, curating the data, and the very fact of them choosing a song and saying, "Hey, I liked that, I will share it with people," would already be a communication.
It needs to be more than that, I want to hear musicianship that has been honed and crafted. The struggle to find their sound. I'm fine with even an amateur musician learning their way around an instrument and being able to put something together that they tracked and mixed.
If a prompt returned the most perfect song, I would still not care to listen as that to me has completely divorced any human element that I would be interested in. Would not find it to be inspiring nor aspirational no matter how "good" it sounded so the models themselves could get exponentially better, but the manner in which it was created will prevent me from ever listening or caring about. It will always be hollow and lifeless.
Again, this is personal preference. If it makes others happy, that's great. In other many other mediums, I'm probably fine with that reduction in human-ness (where others may not be).
Fine, but you've now established this loop where one must find and analyze the human struggle in the music before qualifying an opinion, how does this jibe with deciding whether or not a tune playing in the grocery store has a catchy beat?
Do you run and grab your phone to id the artist before you decide to tap your foot?
> Although, even in that case, regardless of the lack of an "artist's intention," there is the interpretation of what people will take from that thing.
Of course. My interpretation is an important part as well, but that comes from me, not the artist, so is a bit different. Well, maybe I should say that the meaning and importance of a song is in the confluence of the artist and myself. I did want to clarify something, though -- I'm not really talking about the "artist's intention" here. That's a different thing, too.
The emotional communication I'm talking about happens even if I have no idea what the artist's conscious intention was, or even if I don't know who the artist is.
> And I do believe a sufficiently advanced AI model would be able to mimic or synthesize human knowledge/worries/dramas in such a profound way that, regardless of "intention to communicate," it would be able to create things that people would relate to and take deeper meaning from.
Perhaps so! But that kind of simulacrum is something I have absolutely no interest in. In fact, I find the idea of it a bit repulsive.
> someone might have used AI, but they were still involved in all parts of the creation process, like writing the lyrics
If an artist actually created the thing, then it's not an AI generated song. It's a human created song that may have involved AI as a tool. I'm talking more about if a human just describes the song they want to an AI and the AI creates the rest.
That said, I'm particularly averse to AI vocals, because vocals are particularly intimate for me. A song that has a machine as a singer is a song I'll reject even if the rest was created by a human.
> the very fact of them choosing a song and saying, "Hey, I liked that, I will share it with people," would already be a communication.
Technically true, but that's nowhere near the kind of communication I'm talking about. That has little value to me unless the person sharing it and myself know each other very, very well. Then, it's a communication/connection between that person and me, which can make it a great thing even if the song wouldn't resonate with me on its own.
I mean, art is inherently about human experience and emotion. Each of us resonates with certain types of art and doesn't resonate with other types. All I'm trying to do here is explore and maybe explain what resonates or not with me. I am in no way saying that anybody else should share my tastes.
If memory serves me, Nano Banana allows generating/editing photos of children. But anything that could be misinterpreted, gets blocked, even absolutely benign and innocent things (especially if you are asking to modify a photo that you upload there). So they allow, but they turn on the guardrails to a point that might not be useful in many situations.
I'm curious about this. I'm not a frontend engineer but enjoy tinkering on simple frontend UIs for my hobby projects, and I've found Tailwind nice for creating encapsulated components more easily. It's funny that it skips the entire cascading part of Cascading Style Sheets though. Are there major downsides besides that?
> It's funny that it skips the entire cascading part of Cascading Style Sheets though. Are there major downsides besides that?
I think cascading is a bad default. It's useful, but only sometimes, and often causes headaches like unintended coupling and confusion about why rules are being overridden. The utility class approach (like Tailwind) makes a lot of issues like this go away. I don't see a good reason why the traditional approach is worth the extra pain or discipline.
CSS can be laser-like specific if you want it to. Want to only affect that thing? Use ids, inline styles or learn how to write proper selectors.
I am by no means a CSS expert, but 90% of CSS issues I heard complaints about boiled down to the complainers not having spent the time necessary to learn the basics. And the other 10% were solved by :has()
That being said, most other styling solutions I had used (e.g. in GUI libraries) will quickly make you wish you had CSS.
> CSS can be laser-like specific if you want it to. Want to only affect that thing? Use ids, inline styles or learn how to write proper selectors.
You've still go all the normal CSS problems like having to debug complex selectors, jumping between files to debug styling, having to name lots of things you're only going to use once, verbose media queries, verbose styling attributes, and not knowing when it's safe to delete styling because you don't know where it's shared.
And for what benefits? To say we're writing CSS "the right way", when it was designed for styling traditional documents and not for complex UIs?
This also strikes me as a "if everyone learned to do it properly, there wouldn't be a problem" statement. It's ignoring the reality that nobody can agree on the proper way to write CSS, and writing discipled CSS is fatiguing and time consuming. And even if you could get everyone to adopt the same approach, the above issues are still a big deal.
It's a real failing when a language or methodology requires you to invent your own complex discipline to tame it (e.g. C and C++), and get everyone on the team to follow this. At some stage, it's better to scrap everything and try again with what was learned to avoid the mess.
> 90% of CSS issues I heard complaints about boiled down to the complainers not having spent the time necessary to learn the basics. And the other 10% were solved by :has()
I really don't agree, I understand traditional CSS and how to use complex selectors and it's just not a good approach except mostly for styling traditional Markdown-like documents and adding your own utility classes. If laser-like specific selectors is something I want more of the time, I want this as the default and for it to be easy.
Tailwind is a very thin layer above CSS and you can't use it properly without knowing CSS. Coupled with the way you reuse styles in Tailwind by using templates (instead of sharing via classes), it solves most of the problems with CSS in a simple way that people find simple to follow.
Tailwind's major downside is it isn't the "traditional" way (which nobody can agree on anyway for complex UI styling), so Tailwind gets attacked for being the wrong way without its benefits and tradeoffs being examined properly.
Jumping between the files? The cascading part means you can always add in a file after and overwrite it. I don't see how this would get better without the cascading part.
My point isn't that CSS is perfect. My point is that someone has yet to show me a better styling language that isn't purely hypothetical. I am happy to learn new languages if there are clear benefits.
I mean having to jump between the HTML and multiple CSS files (which often you have to track down by using the browser inspector) to make edits to while styling things. When HTML and styling are tightly coupled anyway and almost always edited together, it just slow everything down for no good reason vs co-locating them together via utility classes.
> The cascading part means you can always add in a file after and overwrite it. I don't see how this would get better without the cascading part.
If you mean overwriting styles set somewhere else, this is what makes CSS confusing and hard to refactor. Cascading is just best avoided whenever you can.
> I am happy to learn new languages if there are clear benefits.
I can recommend looking at Tailwind. Make sure to use it with some kind of templating language e.g. so a "button" component goes in a template file, as that's the way you reuse styling (vs copy/paste) which critics seem to miss. It makes styling much simpler and quicker (everything is co-located, no need to write selectors, no need to make up class names, very concise syntax especially for mobile), especially if you're doing complex responsive designs.
I am editing my CSS files predominantly in the browsers inspector this has the benefit showing you what it looks like directly, potentially even with mobile preview.
I also don't need to hunt down CSS files because I rarely ever got more than 5 css files in projects I authored myself usually more like three: style.css for general stuff, fonts.css for fonts and multiple foo.css for page/section-specific stuff that isn't needed elsewhere thus only one of those is ever loaded.
For most "theming"-like stuff I make extensive use of css variables I want all my things to look like the button? Well just add the same styling with the variables in place.
As for the divide between HTML and CSS: to me HTML is 99% semantic. That means I describe the information as it should be described and the rest is done in CSS. Nowadays I rarely ever feel the need to create useless divs or go back and edit the HTML to fix styling issues thanks to grid layouts.
It wasn't always that way, but thst was to a degree the point of the linked article as well.
> As for the divide between HTML and CSS: to me HTML is 99% semantic. That means I describe the information as it should be described and the rest is done in CSS.
How do you visually group and align elements in designs like this one by only using semantic elements? I agree there's a few things in there that could be ol/ul/li tags but there's a lot of divs that are just flex containers or flex items. I don't see what useful semantics they could all have when they're mostly for cosmetic reasons.
Those warnings are stupid.
reply