@pvncher Any good review swarm skill you have to share please? I'm still using the one from @Dimillian, wanted to know if there's something more up to date!
@simpsoka App wise: You guys have to lock in the split between Windows & WSL2. Why would the app choose to create a worktree in Win when a user selected WSL2 as the workspace? It's been like this since release iirc.
you know it's rough when sama and dario are using the same language: working for general use and access again
hopefully we're not sealed in the permanent underclass for too long
if it'll be restricted, how do you see local AI becoming viable? The amount of compute necessary to use in any viable and meaningful way open source models, "close" to frontier closed sourced models, is impossible atm. Sure we can run a 30B model etc, what about the frontier in the hands of the people though
I'm using the Codex app on Windows (I know) with WSL2; I knew I was paying the Windows tax, but man... The length of tasks was absolutely horrible. 10 minutes to make a commit of what you did? What are we doing
Went back to CLI, everything is much smoother and faster. And also the amount of time being spent in "Thinking" in the app, doing nothing... There needs to be some fixing for this
@LexnLin Can we acknowledge the absolute dog water speed of 5.5 nowadays finally? Like I said before, a lot of it is spent "Thinking", and in the end you're just in a load balancing queue
Nowadays I have a strong feeling that "Thinking" is used as a load balancing/management, and actually does nothing but putting you in a queue to wait until the model can continue working.
Length of tasks seem to have dragged on way longer then needed.
Hopefully 5.6 is more efficient
@NoahShreve are you only using webflow? (what about nextjs & others; clients possibly asking for a codebase instead of a webflow?)
what are your retainers looking like and how do you manage to sell them?
@thsottiaux Tibo I hope you guys aren't hiding streaming issues behind "Thinking". Noticed streaming/reconnect issues last week, and it seems that they're now happening in the Thinking. The latency is huge between sending a message and the model actually showing work being done
people won't answer as they used to by text
they'll answer in ways they never did before
they'll question you in ways they never did before
and it's horrible
just let me speak straight to claude, it'll be easier for both of us lol
You can tell when people use and talk to Claude all day long. It's parasitic and changes the way people think
Just be trained for engagement or something. Seriously horrifying. 4o nouveau
@banteg I'm tempted to get an MBP because the toolchain for a lot of products is just better there. But the trade off is less attractive when you have a beast machine with W10/11 + WSL2; best of both world for working, and being able to play video games