I'm surprised that the norm isn't to do many of them, that someone hasn't written some software package that just "aggregates" multiple platforms on the client side.
I mean, if I were running a business and wanted to have a social media presence, that's probably what I'd want to have.
You'd expect it -- most online companies do, as it makes sense for companies with high fixed costs and low variable costs -- to have a growth phase, during which it loses money but aims to grow by being very appealing. Once it's grown as far as it reasonably can or as money permits, the growth phase ends and the monetization phase begins. Twitter's growth phase was over. It would never have been expected for Twitter to just lose money forever.
According to the article text:
A key driver behind the success of Chinese apps is that they have integrated e-commerce into their platforms, blending entertainment and networking with sales to monetize their famously addictive algorithms, according to Chinese social media experts and marketing firms.
Isn't Jenkins a continuous integration system? I'd think that you'd want a configuration management system (like ansible) if you're not trying to set up CI infrastructure.
They don't really have a choice. TV viewership has been on decline.
kagis
https://www.oberlo.com/statistics/how-much-tv-does-the-average-american-watch
Time spent watching television: digital video vs. traditional TV
From 2021 to 2025, daily time spent watching traditional television is projected to plummet by 28 minutes. In contrast, Americans are spending increasingly more time consuming digital videos. In 2021, they spent an average of three hours a day enjoying digital videos. By 2025, this is set to grow to four hours, marking an overall rise of 33.3%.
I suspect that there are many websites that already dynamically generate an unbounded number of pages based on the links one clicks, and that Web spiders will have needed to deal with those for as long as there have been people spidering the Web, which is going to be no later than the first Web search engines.
I'd guess that if nothing else, they cap how far they spider a site. Probably a lot more sophisticated, use heuristics to figure out which sites are more worth spending indexing resources on, as it's not just whether to spider but also the frequency with which to do so. Some parts of a site are more "valuable" than others -- for a search engine, a more desirable target for users clicking on results -- and some will update more frequently and are more-useful to re-spider at higher frequency. Google will return current news articles, yet still indexes a large portion of the content out there. They won't be doing that by simply sending GoogleBot at everything that they've indexed at a fixed frequency.
Isn't that 50 years later?
SDL is a widely-used-on-Linux platform abstraction layer. A lot of games have targeted it to make themselves more portable and provide some basic functionality.
If you've been on Linux for some years, you've probably run into it. I'd guess that most Windows users probably wouldn't know what it is, though.
Anthropic chief says AI could surpass “almost all humans at almost everything”
Agreed.
shortly after 2027
I doubt that.
I think that there are near-term things that have real value that one can build with existing approaches. However, I think that we are still a fair bit of research away from beating humans across the board.
Three years is "build a software package when you already know what you're doing", not "start on a grand research program" timescale.
Ditto, though the LoRA might explain it.
The base model, Pony Diffusion, doesn't normally act like that in my experience. Maybe if you generated an enormous number of images and culled them.
The simplicity of the style might also help -- less to change.
breathtaking cat in neon paints . award-winning, professional, highly detailed
Negative prompt: signature, text, watermark, ugly, deformed, noisy, blurry, distorted, grainy
Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 7, Seed: 3, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.16, Token merging ratio: 0.5, Token merging ratio hr: 0.5, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+, Version: v1.7.0-133-gde03882d
city, diamine art, shimmering
Negative prompt: bottle, photograph, text, signature
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 0, Size: 2560x1440, Model hash: ebf42d1fae, Model: realmixXL_v15, Version: v1.7.0-133-gde03882d
A couple notes:
I'm aiming for a fountain pen look; you can do quite a bit with fountain pen ink, creates a lot of color gradations and so forth, though in real life, the ink is hard to control. I really like the look of the stuff. It's kind of like more-elaborate watercolors. I'd spent some time in the past unsuccessfully trying to get such a look with "fountain pen" and similar terms, and didn't get satisfactory output; I got pictures of fountain pens, but not much that looked like a fountain pen artwork. I finally hit it by trying specific ink names; the one here is reference to a line of a "shimmering" inks made by Diamine.
This image was generated natively at 2560x1440; apparently, at least with --medvram, this is possible on a 24GB video card. Automatic1111 does not, by default, permit a user to create images larger than 2048 in any dimension; typically, users upscale to these resolutions. However, one can edit ui-config.json directly and modify txt2img/Width/maximum to be higher numbers and it will work, as long as there is enough video memory.
Stable Diffusion tends not to do so well generating images much larger than the training size; what I expect happens is that it starts to converge on different images in different parts of the large image, and doesn't wind up having the image as a whole converge. I would guess that it's possible to tweak the ancestral noise settings so that there's enough noise added at each stage to bump it out of whatever local minimum it's converged on, but at least with the standard settings, this isn't really possible. This means that one tends to get the sort of "distorted monster" look with lots of people merging into one. I ran through a couple different types of scenes, looking for something that wasn't too-badly impacted; I'd noticed before that landscapes tended not to be too badly impacted, as Stable Diffusion could reasonably fill in, say, a cliff face between two existing cliff faces that have been converged on in a way that it can't fill in when two different human faces that collide with each other have been converged on in different parts of the image. Cityscapes also seem to do all right; SD can fill in similar buildings, fit things together pretty well. Basically, one wants a scene that doesn't have giant features that can't reasonably be reconciled with each other.
An image of Santa Claus and Krampus.
This was a two-step generation. First, creation of a 1280x720 image. Three columns were used in Regional Prompter; one column in the center that used both Santa and Krampus as a negative prompt to maintain some separation between Santa and Krampus in the image:
gouache illustration
ADDCOMM
santa claus
ADDCOL
ADDCOL
krampus
Negative prompt: woman
ADDCOMM
ADDCOL
santa claus, krampus
ADDCOL
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 1, Size: 1280x768, Model hash: ebf42d1fae, Model: realmixXL_v15, RP Active: True, RP Divide mode: Matrix, RP Matrix submode: Columns, RP Mask submode: Mask, RP Prompt submode: Prompt, RP Calc Mode: Attention, RP Ratios: "1,.5,1", RP Base Ratios: 0.2, RP Use Base: False, RP Use Common: True, RP Use Ncommon: True, RP Options: [False], RP LoRA Neg Te Ratios: 0, RP LoRA Neg U Ratios: 0, RP threshold: 0.4, RP LoRA Stop Step: 0, RP LoRA Hires Stop Step: 0, RP Flip: False, Version: v1.7.0-133-gde03882d
The Hires fix upscaler built into Stable Diffusion doesn't work with a regional prompt, as each tile it works on will have the regional prompt applied to it. So it's not possible to generate a full-resolution image in a single step.
I took the image and 2x upscaled it to 2560x1440 as a separate step in img2img using the Ultimate SD Upscaler extension:
gouache illustration
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 824048783, Size: 2560x1536, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.16, Ultimate SD upscale upscaler: SwinIR 4x, Ultimate SD upscale tile_width: 512, Ultimate SD upscale tile_height: 512, Ultimate SD upscale mask_blur: 8, Ultimate SD upscale padding: 32, Version: v1.7.0-133-gde03882d
That produced the final image.
rooster, neon paints, black background
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 15, Size: 1024x1024, Model hash: ebf42d1fae, Model: realmixXL_v15, Version: v1.7.0-133-gde03882d
dia de los muertos, sailor moon
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 7, Size: 1024x1024, Model hash: ebf42d1fae, Model: realmixXL_v15, Version: v1.7.0-133-gde03882d
A series of famous present-day landscapes in the US picked from this list rendered in the style of John William Casilear, a 19th-century American landscape painter. They don't look quite like the actual locations -- to do that, I expect that one would probably do better to start with a photograph and do img2img transforms -- though they do get new, interesting landscapes with the feel of them.
Rendered in Stable Diffusion.
These are 2560x1440, aiming for a full-screen image on a 1440p screen, and are probably suitable for use as desktop backgrounds on such screens. If you are viewing these on a desktop in 16:9 aspect ratio (the most-common these days), you can probably benefit from opening the images and viewing them fullscreen. I include a link for each, as the lemmy Web UI doesn't have a great way for a user to view an enlarged inline image (IIRC the kbin Web UI does).
Scotts Bluff, by John William Casilear
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 12, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent, Version: v1.7.0-133-gde03882d
Three Sisters Springs, by John William Casilear
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 6, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent, Version: v1.7.0-133-gde03882d
Kent Falls, by John William Casilear
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 19, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent, Version: v1.7.0-133-gde03882d
Avenue of the Giants, by John William Casilear
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 18, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent, Version: v1.7.0-133-gde03882d
Buffalo National River, by John William Casilear
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 10, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent, Version: v1.7.0-133-gde03882d
Denali, by John William Casilear
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 15, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent, Version: v1.7.0-133-gde03882d
Dismals Canyon, by John William Casilear
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 13, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent, Version: v1.7.0-133-gde03882d
Santa Elena Canyon, by John William Casilear
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2, Size: 1280x720, Model hash: ebf42d1fae, Model: realmixXL_v15, Denoising strength: 0.7, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent, Version: v1.7.0-133-gde03882d
I usually use the Euler a sampler; I find that the ancestral samplers ("-a") tend to do better on things like fingers (though they have the technical drawback that adding more samples also alters the image; you can't just "converge" on a single image and throw the minimum number of samples required at the problem). For at least these painted landscapes, I didn't seem to have any visible issues using a non-ancestral sampler.
I tend to render at 1024x1024, the SDXL training resolution; this helps to avoid things like people with extra limbs. For landscapes, this doesn't seem to be an issue for me, and using a different aspect ratio seems to work without flagrant visible issues. My experience is that it can lead to similar elements replicated in an image, but doesn't seem to be a crippling issue for these landscapes.
One of the largest issues I ran into was reflections in water that didn't quite match the image above the water; I tended to select images without reflections in the water, or at least ones where the issue was less-obvious, though having reflections in water is common in landscape paintings and in Casilear's actual works. My impression is that Stable Diffusion can deal well with situations where a perfect mirror of an image is expected, but not with one that is distorted in various ways -- as is the case for water with ripples.
I usually post images in PNG format. I use JPEG here; they are less than a tenth the size for images in this style, and I feel like doing PNG for the larger images I'm doing here places unreasonable load on the lemmy host I use, which is -- at no charge -- hosting posted images for users of the server. I wasn't able to see any artifacts in visually-inspecting PNG and JPEG versions.
I rendered batches of 20, and picked the output that I subjectively felt was most-aesthetically-pleasing.
Prompts and other information to reproduce the images are included for each.
cat,white ink,black background, (red eyes:.3), (uncanny wide smile:2)
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 0, Size: 1024x1024, Model hash: ebf42d1fae, Model: realmixXL_v15, Token merging ratio: 0.5, Version: v1.5.1
(barack obama:.5),running,blue,orange,red, by Frank Miller,style of sin city
Steps: 40, Sampler: Euler a, CFG scale: 7, Seed: 0, Size: 1024x1024, Model hash: ebf42d1fae, Model: realmixXL_v15, Token merging ratio: 0.5, Version: v1.5.1
This is really to showcase another neat phenomenon I've run into -- I didn't spend time trying to find the coolest image I could find.
I really like hard-light, black-and-white pen-and-ink drawings (as, with the above example, Frank Miller did with Sin City). And Stable Diffusion can do a pretty good job of imitating those styles. However, those can be a little intense, very high-contrast, be a bit overwhelming to the viewer.
It looks like, if one adds a few color names to an image, Stable Diffusion winds up starting with the black-and-white art style, but will then fill in limited amounts of color...and does so in reasonable places.
Here's the same prompt without the color terms:
(barack obama:.5),running, by Frank Miller,style of sin city
Steps: 40, Sampler: Euler a, CFG scale: 7, Seed: 0, Size: 1024x1024, Model hash: ebf42d1fae, Model: realmixXL_v15, Token merging ratio: 0.5, Version: v1.5.1
A few other examples, starting with black-and-white styles, and using the same prompt:
(barack obama:.5),running,blue,orange,red, by Alex Toth
Steps: 40, Sampler: Euler a, CFG scale: 7, Seed: 0, Size: 1024x1024, Model hash: ebf42d1fae, Model: realmixXL_v15, Token merging ratio: 0.5, Version: v1.5.1
(barack obama:.5),running,blue,orange,red, by Steve Ditko
Steps: 40, Sampler: Euler a, CFG scale: 7, Seed: 0, Size: 1024x1024, Model hash: ebf42d1fae, Model: realmixXL_v15, Token merging ratio: 0.5, Version: v1.5.1
These aren't ideal images -- it'd probably be better to reduce the "barack obama" weight further so that the more-detailed, photographic appearance doesn't noticeably come out in the final image on the face, something I've observed happening with a number of images of politicians. But they do a pretty good job of demonstrating the effect.
Yeah, this kinda bothers me with computer security in general. So, the above is really poor design, right? But that emerges from the following:
Writing secure code is hard. Writing bug-free code in general is hard, haven't even solved that one yet, but specifically for security bugs you have someone down the line potentially actively trying to exploit the code.
It's often not very immediately visible to anyone how actually secure code code is. Not to customers, not to people at the company using the code, and sometimes not even to the code's author. It's not even very easy to quantify security -- I mean, there are attempts to do things like security certification of products, but...they're all kind of limited.
Cost -- and thus limitations on time expended and the knowledge base of whoever you have working on the thing -- is always going to be present. That's very much going to be visible to the company. Insecure code is cheaper to write than secure code.
In general, if you can't evaluate something, it's probably not going to be very good, because it won't be taken into account in purchasing decisions. If a consumer buys a car, they can realistically evaluate its 0-60 time or the trunk space it has. But they cannot realistically evaluate how secure the protection of their data is. And it's kinda hard to evaluate how secure code is. Even if you look at a history of exploits (software package X has had more reported security issues than software package Y), different code gets different levels of scrutiny.
You can disincentivize it via market regulation with fines. But that's got its own set of issues, like encouraging companies not to report actual problems, where they can get away with it. And it's not totally clear to me that companies are really able to effectively evaluate the security of the code they have.
And I've not been getting more comfortable with this over time, as compromises have gotten worse and worse.
thinks
Maybe do something like we have with whistleblower rewards.
https://www.whistleblowers.org/whistleblower-protections-and-rewards/
So, okay. Say we set something up where fines for having security flaws exposing certain data or providing access to certain controls exist, and white hat hackers get a mandatory N percent of that fine if they report it to the appropriate government agency. That creates an incentive to have an unaffiliated third party looking for problems. That's a more-antagonistic relationship with the target than normally currently exists -- today, we just expect white hats to report bugs for reputation or maybe, for companies that have it, for a reporting reward. This shifts things so that you have a bunch of people effectively working for the government. But it's also a market-based approach -- the government's just setting incentives.
Because otherwise, you have the incentives set for the company involved not to care all that much, and the hackers out there to go do black hat stuff, things like ransomware and espionage.
I'd imagine that it'd also be possible for an insurance market for covering fines of this sort to show up and for them to develop and mandate their own best practices for customers.
The status quo for computer security is just horrendous, and as more data is logged and computers become increasingly present everywhere, the issue is only going to get worse. If not this, then something else really does need to change.