@JackEllis @verymadpanda Since we switched to CloudFront, we actually have seen a couple gateway errors from time to time. They say it's normal, and looking at their nines (99.9%), it's within the "expected error rate". Still not great for the user(s) though.

@RozenMD I generally find writing custom SDKs to be the only solution when the original one doesn't have any tree shaking or is severely outdated. Pulling in all of their APIs when you only need one or two endpoints is really wasteful (usually those tend to use classes for some reason).

@MikeLewis @leesimin Love to have shortcuts - but there seems to be an issue with the 3 buttons turning totally white when used with the new tinting option for app icons (i tint mine pure red during sleep mode).

@JackEllis (can't edit, but probably obvious: i meant non-native speakers)

@Kevin2Chambers we are very happy with @CustomerIO (data hosted in the EU, their parcel email design system as well as the way they allow attributes on people, objects and relationships)

@JackEllis Unexpectedly had the same happen to me this year. It's been a (sometimes painful, but often amazing) blast so far!

@JackEllis Yup! it was/is a very old house and they're not allowed to change out the windows as it is landmarked/listed.

@JackEllis Can confirm: some time ago, I had heavy nose bleeding and very dry skin just by being in my room and decided to test the humidity: <20%. got myself one of those air humidifiers and have kept a good eye on it since!

@JackEllis I actually love the Claude artifacts for really small, bespoke one-off things I wouldn't want to put the time in to code myself. Either they're tools for myself, or simple prototypes of ideas to help during ideation. They're a neat timesaver (but the scope has to be very small).

@AlanZucconi I wonder if they can/will make distinctions about what kind of AI is used to train on the data. You probably need to have an AI trained on copyrighted materials to see if someone is posting infringing material online (as a moderation layer), but not for all cases...

@lucasmeijer To me, it's way worse than o1-preview :( You could one-shot the preview model with complex tasks, and with o1 you're back to a conversation because it messes up left and right (coding task, instruction following etc.).

@adamdangelo @xanderatallah We currently aren't using AI judging just yet because we found that there aren't really enough models out there that, surprisingly, actually CAN judge writing style. When asking if a scene uses "show, don't tell" most say yes, even if professional editors would mark it as a fail.

@OfficialLoganK ping! I've sent a mail about being unable to use the /models OpenAI-compatible API with an AI Studio API Key a while back. It always returns an unauthorized error, while generating text works just fine (with the same key). Not sure if you saw it yet :)

@JackEllis I've played with it in the past (fargate), but I like the simplicity of Lambda. Easy to roll back or roll out staggered releases, too.

@JackEllis We're using provisioned currency already, but it feels bad paying for waiting for the DB/IO. Average requests are at 20ms, but when one bigger request hits, the full provisioned lambda blocks for a while :/ ECS would solve this, right?

@lucasmeijer not sure if @SmartcatAI can do it? I know they do video + audio at least

@adamwathan noticed that on iOS! We had to make hover a fixed color instead :(

@itchio Will there be more nuance to this? like does running handmade assets through AI upscalers count as AI generated content? Just curious.