Leonidas

From chemical to software engineering.

Founder of:

[Leonidas @Leonidas76167]

@ptrschmdtnlsn @tmkadamcz The api probably does, but keeping the client side state in sync with the true server state is another task.

[Leonidas @Leonidas76167]

@grhmc They charge for egress and ingress, of course they dont :D

[Leonidas @Leonidas76167]

@MansiCodez keep it simple: add a dedicated table for the 100 top scores and update that when a new score gets added. i would try to use my existing database / only add redis if verified its truly needed perf wise

[Leonidas @Leonidas76167]

@vanilagy @wiedymi That page performs very shitty, it would benefit greatly from virtualization

[Leonidas @Leonidas76167]

@dhh Is Omabot a new tool you built or just the nickname for an existing agent \ harness?

[Leonidas @Leonidas76167]

@valcoholics1 On macos its hidden under more actions or sowmthing called lile that

[Leonidas @Leonidas76167]

@0xSqueemo @Jonathan_Blow I dont see the original post anymore, i think it stated 32bit hashes - so its ~ 1 in 4 billion risk

but hopefully leans implementation uses a algorithm where it first checks hashes - and if they are even compares values.

that would make for a sane optimization

[Leonidas @Leonidas76167]

@0xSqueemo @Jonathan_Blow i get that hashes are easier to compare - but still: in mathematics a proof has to be true without any doubt.

So a hashing comparison is wrong. for faster feedback during development hashing can be used, but the final compile should not sacrifice correctness for speed

[Leonidas @Leonidas76167]

@0xSqueemo @Jonathan_Blow ? Why should you have to use them. Collisions can occur, doesn’t seem fit to proove a theorem for mathematics

[Leonidas @Leonidas76167]

@thdxr @tomerweller will the idle ram footprint be reduced?

currently each opencode serve idles at around ~400mb - which for a single user isnt bad / a problem,

but for a central server that serves multiple team members (each having their own opencode server running) not ideal.

[Leonidas @Leonidas76167]

@DanielGlejzner For solid to perform well the data strcuture of the app has to be broken into signals. Reacts mental model is often nicer to work with when having large datasets (large in the sense of alot of schemas) synced into a dedicated store and rendering that to ui at different places

[Leonidas @Leonidas76167]

@valyala And for more obscure databases you can always resort to docker

[Leonidas @Leonidas76167]

@valyala What do you mean with production databases dont run under macOS? Postgres & Mysql run fine, setup is as easy as on linux

[Leonidas @Leonidas76167]

@aarondfrancis At this point you can use the migration assistent :D

[Leonidas @Leonidas76167]

@yashmp2004 I go with swift, which is actually a very great programming language

[Leonidas @Leonidas76167]

@thegenioo @blueemi99 Well but it takes very long to finish a task and uses a lot of tokens. From my few examples luna is the better model for dev adjacent work

[Leonidas @Leonidas76167]

@kimmonismus Hmm i dint know if its that good of a catch - just watched deepseek flash waste 200k tokens on adding a small ui feature - to add a dropdown for some links if space gets too small and fail at it. Opus 5 literally took a few seconds for the same

[Leonidas @Leonidas76167]

@kimmonismus Sure this is only posttraining? The model size and pricing chaned, seems unlikely

[Leonidas @Leonidas76167]

@vanilagy Well thats not a giod workaround for data heavy apps sadly

[Leonidas @Leonidas76167]

@allgarbled Which profession does this better? From my experience every field tends to use jargon that often get in the wy of beginners. In computer science the information atleast is nicely digitalized so googling is easy