@aimalysheva I always felt that time estimates during the planning phase were partly shaped by labs to create a sense of how much work was being done. Not necessarily that the estimates were wrong, but they never felt fully genuine.
@jangiacomelli@samuelcolvin If you are using Claude Code then switching between models should be trivial, where is the friction coming from? Is there some setup that works with these models that doesn't work with Fable?
@EmpireEnjoyer3@ByronYork So you think the only way the US could have avoided a military defeat was to commit war crimes? Even then this would not result in the fall of the regime.
@mitsuhiko@badlogicgames@Steve_Yegge A hypothesis around is that is style has been introduced because of using cyclomatic complexity in the RL reward term. The models tend to extract helpers or use polymorphism instead of branching aggresively.
@Loc0m0 اگر قبلا با Claude کار کرده باشی معمولا یک CLAUDE.md ذخیره میکنن و codex این رو نمیخونه مگر این که به عنوان fallback توی کانفیگش اضافه کرده با��ی. از این نظر گفتم که اگر خواستی مقایس�� کنی باهم برابر باشن.