@hunvreus@pidotdev a mailbox mcp between claude code + codex + opencode for local models is what I roll with. Opencode and codex both opensource + app-server. Haven't bothered with reusing claude code or codex with different models, but it's not difficult. CC remote app is nice for top-lvl control
@goshatch Macros not dealing with that position in an RL environment might have pushed it that way. Although claude doesn’t seem to have this problem. Partially because it usually has my code in context where the string usually does go before the vec.
@unconed Yeah CRDTs are very limited in scope. Had a look at replicache for a general solution, they also use fractional indexing for undo. But basic primitives like locks can go a long way. Would really like to read more about your views on data modeling and api design!
@ovstoica@turjanofmiir@shipclojure Meh, mount was good, to me it’s just another layer of indirection and encouraging overengineering for what benefit? Allowing bypassing of namespace load order restrictions that are there for a good reason. I don’t get the whole “component/system” thing people talk about.
@ovstoica Core.async is more flexible, but it’s easier to lose track of how data flows as you put them on a channel and don’t “know” what happens after that.
@ovstoica I like that it has you compose a program with all the resources a feature uses, is elegant with regular clojure and coroutines but still has a good way to handle errors. Lets you think about back-pressure and has a pretty minimal, but sufficient amount of abstractions to learn.
New paper: The Platonic Representation Hypothesis
In which we posit that _different_ foundation models are converging to the _same_ representation of reality.
paper: https://t.co/z0rh2gPeCc
website: https://t.co/Qo3BLTo25I
code: https://t.co/T0YUxG77mg
1/8
Finally, to conclude our technical programme we have Donald C Jackson from Stanford presenting "A System for Measurement and Analysis of Aircraft #Noise Impacts". #OSN21