Timotius Sitorus

Web Developer. Hackathon Hacker. Peak GM @TFT Player.

[Timotius Sitorus @timsitorus]

@aidenybai Could totally see this being run as a CI step or integrated with PR agents like @greptile

[Timotius Sitorus @timsitorus]

Finally looking at lower end LLM models and I'm really liking @Kimi_Moonshot K2.5.

That being said GLM 5 and Minimax m2.5 seems to be ahead in benchmarks but I can't seem to get good outputs consistently. Anyone got tips?

[Timotius Sitorus @timsitorus]

@bekacru It seems like you're already in the unjs ecosystem with rou3. Was there any reason you didn't just use h3?

[Timotius Sitorus @timsitorus]

@pizonka662 Also I fail to see why Augment stats removal would effect comp choices unless theres strong correlation between the comp and the augment, in which case you would easily be able to make that connection, no stats needed.

[Timotius Sitorus @timsitorus]

@pizonka662 But the same could be said about stats. If you saw that ashe-sett vanq had good stats in China and you tried it without studying why its good, you would still get 8th.

[Timotius Sitorus @timsitorus]

@mfts0 @vercel @paw_lean @rauchg Would love a more detail write up on this!

Ping @theo this might be a good video idea

[Timotius Sitorus @timsitorus]

@rexan_wong @vercel Before you actually consider moving off Vercel, try analyzing what is causing you to go over the 1TB data transfers. 30k users shouldn't be enough to go over that cap if you're doing everything correctly.

Also spending caps. You should have this set.

[Timotius Sitorus @timsitorus]

@ankur_datta_007 @prisma Pulse error is gone for now. Region is Asia Pacific.

[Timotius Sitorus @timsitorus]

Checking out @prisma Postgres for a weekend project and suddenly the db stops responding and Pulse errors out (even though I'm not using pulse).

Anyone else experiencing this?

[Timotius Sitorus @timsitorus]

@boristane Does a single worker have some form of replication to handle concurrent requests from both the fetch and queue?

[Timotius Sitorus @timsitorus]

@boristane How do you keep track of all the durable object that have been spun up from all the errors from a single client? Is there a db/durable object that tracks the durable objects that represents errors from a single client?

[Timotius Sitorus @timsitorus]

@Cohn_Jarmack @ethanniser @RhysSullivan @ksw_arman You can use an online tool or make a script that uses Sharp (which nextjs uses) to do this. Then upload them to s3 or another hosting service to serve them. There are also services like cloudinary that do this but it might cost some money.

[Timotius Sitorus @timsitorus]

@BraydenWilmoth @Cloudflare Does cf workers have a way to replicate themselves to prevent overloading? Because in that architecture you would have single point of failure in your core service worker.

[Timotius Sitorus @timsitorus]

@Cohn_Jarmack @ethanniser @RhysSullivan @ksw_arman Should probably look into optimizing images. Loading takes up most of the wait time. NextMaster does it using next/image but you can probably just transform them ahead of time.

[Timotius Sitorus @timsitorus]

@jmwind @vercel @typesense @supabase What was the bug that caused the increased vercel bill?

[Timotius Sitorus @timsitorus]

@threepointone Maybe some sort of built in cookie banner component since @rauchg was complaining about it

[Timotius Sitorus @timsitorus]

@ethanniser @nextjs @RhysSullivan @ksw_arman Correct me if I'm wrong but from your scripts it looks like you're generating 512x512 images but it looks like the maximum product image size is 256x256. Is there any reason why you didn't just generate this size to begin with?

[Timotius Sitorus @timsitorus]

@btn0s @rauchg Yup straight up sprite. Learned about it from @wesbos's video on it

[Timotius Sitorus @timsitorus]

@BraydenWilmoth Have you done anything to deal with the 1000 req/s limit of DOs? This is currently the biggest deal breaker for me from investing more into DOs