Counterpoint: I think they usually look dumb, but seeing them does not prejudice me for or against the text of your article.
To clarify, I do sometimes get mad at actual photos that are poorly chosen, and the same is true of AI-generated photos demonstrating bad art direction, bad taste, or bad judgment on the part of the author or editor. But them being obviously AI-generated has nothing to do with it.
> I know there’s plenty of things you can roast my blog for but at least you know for a fact you’re getting the thoughts of a real human being and not some LLM.
Why? I'd venture that a plurality, if not a majority, of people who can write well cannot create good digital art, and vice-versa. The use of AI images on a blog post, assuming you're there for the written content, seems at best a very weak signal of any kind.
Nah.. it means you're cool steeling others creativity.
It's a strong signal that your writing from the POV of a person who doesn't object to stomping on the shoulder of giants.
If you don’t have an image that adds useful information to your document, don’t add an image at all. People on slow or metered connections will thank you.
Fair point. You can't know for sure that I don't use LLMs for my blog. I think it would be obvious it's a real human being if you read a few blog posts.
It was novel and funny when it first came out. Now it's not novel and just crap. Its fine to have no image on your blog post, its better than a slop generated pic.
Agreed. I still think it's cool from a technology point of view. But it doesn't inspire confidence when I see these generative images in blogs/articles.
I spent weeks playing with Midjourney until 3am. I also can’t stand AI generated pictures anymore.
Something about it being so generic and not making sense really triggers the uncanny valley sense now. If anyone is not sure what I mean, ask an AI to generate a map of your home country (or better - your metropolitan area) and you’ll see exactly why it’s so off putting
Embarrassingly, I spent even longer with Midjourney to come to the same conclusion.
It really isn't an issue with AI art but the limited, uncreative output of AI art models. I suspect many artists have been removed from earlier midjourney models too as well along with human input has softened the edges overall.
Then is just a matter of time that someone will comment on how you have to use stable diffusion and how much better stable diffusion is but that is because they mean "art" in the context of AI hentai.
Photoshop AI fill sounds like the greatest thing ever but the output is the same useless slop, not even as good as Midjourney. Only good for press releases and cherry picked best examples.
I don't like AI art in blogs because to me it is just signalling you haven't spent enough time with AI tools yet to understand these images look like shit. The other blog content then is framed in the same light.
I stopped using AI-images as social media thumbnails since a) they take a surprising amount of effort in order to make distinct and non-generic with techniques such as ControlNet [e.g. https://minimaxir.com/2023/03/new-chatgpt-overlord/featured....] and b) thumbnails don't matter anymore for personal blogs since social media sharing of links is dead.
I'm not sure if it's the whole social media, but X (formerly Twitter) now discourage you to share links. Yes, apparently attaching a URL will rank you lower in algorithm.
Earlier this year I was visiting friends and one of them was watching some terrible YouTube videos full of ancient alien type nonsense and full of AI stock footage. Half of it didn't make any sense, and it was just a bizarre experience - mixed up nonsense blather illustration with nonsense imagery.
The friend who was watching this stuff is not the brightest, and he's extremely gullible. I'd rather he just watch spiderman or something, he'd probably be better informed about the world.
> Earlier this year I was visiting friends and one of them was watching some terrible YouTube videos full of ancient alien type nonsense
He didn't need to do that - there's some really great videos full of ancient alien nonsense!
[EDIT - I actually like the ancient alien nonsense. It's no less credible than the various spirituality, religious, mystic or occult beliefs held by 90% of the planet. If you're judging someone's intelligence on whether they like to watch ancient alien theories/spiritual/religious/mystical/occult shows, then you're so disconnected from humanity that your opinion probably doesn't matter to anyone, anyway.]
My problem with the AI-generated images that most people use for their blogs isn't that they're AI-generated, it's that they're bad. People use, I think, mostly DALLE-3 to make these super-busy infographic type illustrations that are completely devoid of meaning because they're AI, full of mangled text and nonsense logos. They're incomprehensible, they maybe kind of convey the idea of "business" or "tech" or something like that but they make no real point.
If you're going to use AI to make a blog image, use it to make something identifiable. If your blog post is about tigers, and you use a picture of a tiger, it doesn't make much of a difference if the picture is AI generated or a stock photo. You didn't take it either way. If the blog post is about your prediction on the next year of tech stocks, use the AI to make a picture of something simple like a computer rather than some kind of Bayeux Tapestry of random tech-like things.
Most AI-generated images I've come across (I have in mind Substacks, Medium posts and other personal blogs) are purely "decorative" & having the same (lack of) purpose as stock photography - e.g. a generic photo of people exercising in an article about fitness. In informational articles I find these pointless & also maybe somewhat in poor taste. But then I'd probably feel the same way about regular stock photography.
There may be potential for AI-generated explanatory visuals though. High-quality diagrams, graphs, map of complex conceptual relationships and so on would be exciting.
That would need to be a whole new model that’s not based on pixels but these nodes and edges. All AI generated maps or diagrams look absolutely unhinged
I agree - at least I take it as a red flag that the blogs content is probably also low effort slob and that I should assign it a rather low credibility
I don't think using AI-generated images is bad. I also don't think using an LLM to help write an article is necessarily bad, either. They're both just tools.
You're equating using these tools with low-effort. And that's often true. But not always.
In our own blog/newsletter (about data and AI engineering), we spend a ton of effort doing a lot of the development work and the research work that goes into posts. Often the post is summarizing an open-source project we've released, sometimes it's summarizing research we've done, etc.
But we're not skilled artists. Nor are we skilled copywriters. And many of us don't speak English as a first language, either. So we use AI to generate images, and we use LLMs to help sharpen the writing on the blog post.
Given a hundred articles (in say, a social media feed) and a finite amount of time and cognitive capacity, how should one decide which articles to read?
If the presentation is AI generated images, and genuinely good content is otherwise “sharpened” with common AI fingerprint phrases in the excerpts and opening titles, what aspect would stop someone from equating it with other “bad” AI generated content and skipping it?
Well, what would stop someone equating it with human-produced bad content? It's not like this is a uniquely AI phenomenon, "bad" content existed three years ago too.
I think you should mostly just not use superficial characteristics like "it includes AI generated art" as much of a signal, and judge it the same way you'd judge anything else - is it from a publication you trust? Is the subject matter of interest? Does the headline and/or skimming the description + first few sentences make it seem interesting?
All of those are better signals of whether it's worth reading something than superficial characteristics. After all, "don't judge a book by its cover" is a pretty common saying for a reason.
I mean, consider not having images, then? Having no images is, IMO, better than having terrible filler images. Most blog posts don’t have images, and the world goes on.
Another thing that makes me close the page as soon as I can press CTRl-W is anime avatars, and if the blog is built like a back and forth discussion between two cartoon characters.
100% with you on the latter point. It's a shame because, so far, where I've seen those, the author does actuallly seem to have some deep knowledge or something insightful to share. But man it's so hard to look past that awful delivery.
To me as the viewer, AI images sometimes pass for real ones, because at a glance they successfully convey a concept. A slight sense of something being off prompts looking a bit deeper and it quickly turns into something akin to a "Spot the difference" game. Spotting one error leads to a chain reaction of finding more errors. Whatever art 'is', from a philosophical standpoint, I must assume it is more than just a medium to convey a concept. At the realization of it being generated, it sort of turns into 'un-art'. Whatever concept is conveyed by an AI image also inherently conveys the concept of "generated".
52 comments
[ 4.5 ms ] story [ 120 ms ] threadBut what I say doesn't matter. What this person says doesn't matter. What A/B tests say matters.
A/B tests say that Ariana Grande is about 20x better than Mozart (according to Spotify stats, actually impressed Mozart holds up so well.)
So something besides popularity in the moment has to matter. Which A/B tests won't capture.
Pageviews. Which is the reason why people put AI images (or any unnecessary stock images) in their blogposts in the first place.
To clarify, I do sometimes get mad at actual photos that are poorly chosen, and the same is true of AI-generated photos demonstrating bad art direction, bad taste, or bad judgment on the part of the author or editor. But them being obviously AI-generated has nothing to do with it.
> I know there’s plenty of things you can roast my blog for but at least you know for a fact you’re getting the thoughts of a real human being and not some LLM.
In fact, I cannot know that.
It really isn't an issue with AI art but the limited, uncreative output of AI art models. I suspect many artists have been removed from earlier midjourney models too as well along with human input has softened the edges overall.
Then is just a matter of time that someone will comment on how you have to use stable diffusion and how much better stable diffusion is but that is because they mean "art" in the context of AI hentai.
Photoshop AI fill sounds like the greatest thing ever but the output is the same useless slop, not even as good as Midjourney. Only good for press releases and cherry picked best examples.
I don't like AI art in blogs because to me it is just signalling you haven't spent enough time with AI tools yet to understand these images look like shit. The other blog content then is framed in the same light.
There's a reason everyone is pushing direct-to-email newsletters.
These social networks are "on the Internet" but they are not "on the Web". The Web is strengthened through linking.
The friend who was watching this stuff is not the brightest, and he's extremely gullible. I'd rather he just watch spiderman or something, he'd probably be better informed about the world.
He didn't need to do that - there's some really great videos full of ancient alien nonsense!
[EDIT - I actually like the ancient alien nonsense. It's no less credible than the various spirituality, religious, mystic or occult beliefs held by 90% of the planet. If you're judging someone's intelligence on whether they like to watch ancient alien theories/spiritual/religious/mystical/occult shows, then you're so disconnected from humanity that your opinion probably doesn't matter to anyone, anyway.]
And what difference do you see between a belief in $RELIGION and a belief in ancient aliens?
From where I am sitting, there is more hard evidence for the existence of ancient aliens than for the existence of any $DEITY who created the world.
If you're going to use AI to make a blog image, use it to make something identifiable. If your blog post is about tigers, and you use a picture of a tiger, it doesn't make much of a difference if the picture is AI generated or a stock photo. You didn't take it either way. If the blog post is about your prediction on the next year of tech stocks, use the AI to make a picture of something simple like a computer rather than some kind of Bayeux Tapestry of random tech-like things.
There may be potential for AI-generated explanatory visuals though. High-quality diagrams, graphs, map of complex conceptual relationships and so on would be exciting.
You're equating using these tools with low-effort. And that's often true. But not always.
In our own blog/newsletter (about data and AI engineering), we spend a ton of effort doing a lot of the development work and the research work that goes into posts. Often the post is summarizing an open-source project we've released, sometimes it's summarizing research we've done, etc.
But we're not skilled artists. Nor are we skilled copywriters. And many of us don't speak English as a first language, either. So we use AI to generate images, and we use LLMs to help sharpen the writing on the blog post.
I don't think that makes our work low effort.
Given a hundred articles (in say, a social media feed) and a finite amount of time and cognitive capacity, how should one decide which articles to read?
If the presentation is AI generated images, and genuinely good content is otherwise “sharpened” with common AI fingerprint phrases in the excerpts and opening titles, what aspect would stop someone from equating it with other “bad” AI generated content and skipping it?
I think you should mostly just not use superficial characteristics like "it includes AI generated art" as much of a signal, and judge it the same way you'd judge anything else - is it from a publication you trust? Is the subject matter of interest? Does the headline and/or skimming the description + first few sentences make it seem interesting?
All of those are better signals of whether it's worth reading something than superficial characteristics. After all, "don't judge a book by its cover" is a pretty common saying for a reason.