masterchief1s1k

Have been passively looking for someone that recommend Visily in these design tool threads, so that I know I'm not alone. It's a pretty solid design tool with tons of AI features that managed to help a technical guy like me make stunning UI. Definitely recommend trying it out.

> Propose to share a proportional number of correctly predicted examples for every incorrect example they come up with?

This is what I have been asked to do the whole time. And yes, like you said it's not very efficient. It would just make them think of the incorrect result as something I need to "improve".

Other team members did shrug it off and simply add those incorrect example as label for the new iteration of model, which did the job and receive high praise. But in my opinion, that just seem like poor AI work ethic.

> The whole thing seems like a communication/political problem, not a technical one, and it's hard to give advice when we don't know the specifics.

Thanks, after thinking about it for a whole day, I guess the correct action is to improve my communication skill, and present the pros and cons of my approach better. I just found out that it is very hard to explain machine learning term without proper visualization.

Thanks for the link. I think it's very helpful, I was able to explain more complex algorithm with visual. Love the explanation on bias-variance and overfitting, I don't think I could get my point across without something like this to visualize my mathematical standpoint.

Instantly downloaded after I see that you have an iPad version unlike IG. Really appreciate it, since I do most of my photo editing on the go with my iPad.

Also gotta say I love the simple and straightforward UI, the dice-rolling thing add a unique touch too.

However, I can see that the app still need a lot of performance tweak, the UX for exiting out of a horizontal image scroll is weird as well, I would prefer to be able to swipe up to exit the view here. Otherwise, cool app thus far!

I would think that since most of their content (photos, stories,..) are optimized for mobile screen size only.

So if their infrastructure doesn't consider much for responsiveness between multiple screen size, I doubt the backend for the web version matter much for them.

Funny that you mentioned Stripe. IIRC the Subscriptions feature can help with the recurrent payment and subscription management out of the box pretty well.

The user management can be a bit tricky through, but I think using their customer portal is also an option, I believed Stripe support enough basic function like CRUD subscription plan, payment method, billing. But this will be highly depended on your use-case.

Maybe you should also consider using the subscription webhooks with your Rust backend for more custom logic.

Deeply agreed, I observed a ton of clickbait and misleading title on Medium these days. Some writers try too hard to catch our attention with fascinating title, only to keep people reading long wall of text that does not seem to be coherent at all. Most of the time, these look like the output of another similar article put through QuillBot.

I guess it is mostly because of their partnership model with writers, because writers are paid monthly based on how much time Medium members spend reading their stories. The longer members read, the more writers earn.

I see that you use euclidean distance of perceptual color and named your color function bad_color_distance. My only input here would be to try something like the CIEDE2000 color difference formula to see if the result can be further improved.

Unfortunately, the problem is the same with many compiled application out there, not just ML.

My rule of thumb in using models from these public model zoo is to always use a separate container instance (or Colab) to check the source code. Then either convert such model to some standardized format like ONNX, which might not be possible for some state-of-the-art models with custom ops afaik.

Or you could just load the torch model first, then load the model state dict separately. Avoid using command like torch.load() since it uses pickle module implicitly, since it is possible to construct malicious pickle data which will execute arbitrary code during unpickling.