I don't understand why they put this up like it's working in their favor. Their website doesn't explain anything extraordinary that makes them different from the average chat app, except that it is europe based.
My thought on this is that LLMs probably mimic writing patterns and structures from quality resources. But they don't construct a plausible thought hierarchy like an average human does, so their train of thought turns into a rollercoaster of thought. So the order of information is for humans completely out of order.
My guess is that it's a known problem, which steered the frontier models into bullet point preference.
Doesn't seem very inclusive. Seems to be another layer to centralize the inbound vulns, gather intelligence and handle them in secret.
It may also turn into another source of pressure. Maybe they manage to sort out the real vulns, but then they come in as high priority to the maintainers.
Many maintainers are already exhausted from their normal work, sans AI noise. Even if they supply fixes, it still requires review.
In best case they could reduce noise but the work is still there. The industry needs to generally fund OS projects to give them the agency to handle it on their own. That's is likely best for quality. If there is still need to filter AI noise then they can add that, but not as a secret opaque thing that controls it all.
Can you elaborate? Do you think the noted issues are non-issues that should be fixed in applications? Which would require a new standard to create stable JSON outputs, which is just one use case that can fail right now.
Unironically QUERY seems to me a "quality of life" feature for the SPA world and the ultimate legitimation to kill link sharing. It is too hard to keep links and app state consistent, so lets just drop it.
I disagree. I think the adoption (or dismissal) of QUERY will show.
First thing that comes to mind is that the idempotency of GET resources are easy to handle. URL's have a fixed size, they can be efficiently hashed, cached and are unambiguous about how they serve this purpose.
It is unclear how the ecosystem will deal with the QUERY requirements. It's easy for apps, but browsers, http caches and servers will take some time to figure out solutions.
Fixing GET would have the same amount of uncertainty in addition to the need to keep current expectations valid. It's not easier, it's harder.
So the meaning of downstream is different for SuperSQL (the direct derivative) and TodoApp (the commercial product)?
Imho, downstream are all consumers/users of that product. I guess it's arguable how a package manager positions in that terminology.
Edit: Also I originally said, "it was the thing used downstream", which just is to describe the consumers. Also not sure what I got into, because the original question is wildly confusing!
I have the feeling that the cloud based providers are just using the freely available segmentation models. It's just speculation, but it doesn't seem to be top priority for them, so they'd just bolt on anything that works.
A problem is also that the cloud solutions need a complex UI to surface segmentation to the user. But the point you have there is that those models are probably not prime time ready yet, surfacing them would actually reveal they are not as powerful as the user expects. Destroying the illusion that AI can just do anything at will.
> At the time this was Stable Diffusion on the backend
The community made models (merges, fine tunes, etc) of that era are all completely overtrained and optimized for portraits and frontal shots. They would try to make a person out of anything. Inpainting faces is already a chore, even with a lot of tooling around that, but inpainting anything else is almost impossible. These models are also especially bad to fit objects naturally into scenes. You can make a crappy necklace or belt work, but introducing a new object into a scene just fails with infinite variety.
They are also much better using 512x512 as resolution, any larger deviation introduces more problems.
Considering you wanted to inpaint banner ads, they would probably get distorted heavily. Those models can't deal with fonts and are bad at a pixel perfect transfers. The only viable way to do this, at that time, would be to manually insert the banner ads and fix the seams with AI. Requires some artistic skill of course.
Your attempt was bold, but with the expectation of just supplying two images and let the models do it, it was impossible.
Well I guess the enterprise side of varnish was quite beneficial to him (and the core devs) over that time. If something gives you financial freedom over 10+ years to work on a thing you love, you can't be mad at all of it.
But if FOSS is deeply encoded into your DNA, you become pretty protective once you realize your project may become extinct sometimes in the future.
> Still, I can't entirely wrap my head around the fact that I live in a world where a machine can create this with minimal intervention by humans, and do a somewhat OK job at it, to the point where I'm willing to spend 10 minutes playing it.
I don't know what that means. I can post something random on social media and there is a chance some person will spend 10 minutes on it. I don't need an LLM nor any money for that effect.