I experienced similar tendencies with Fable as well.
Even in fresh sessions with minimal context like a file with a couple hundreds of lines of code, it would frequently ignore clear instructions, avoid work, and even sometimes “think” things like “looking for ways to code without approval”.
The model is just tuned for long running and doesn’t like to work with a human in the loop, or work back and forth.
Even in fresh sessions with minimal context like a file with a couple hundreds of lines of code, it would frequently ignore clear instructions, avoid work, and even sometimes “think” things like “looking for ways to code without approval”.
The model is just tuned for long running and doesn’t like to work with a human in the loop, or work back and forth.