Writing
Have you ever tried to teach an LLM to balance a stick with words?
Using a tiny LLM to balance an inverted pendulum with words.
Have you ever tried to teach an LLM to balance a stick with words?
Of course not because you are a sensible person.
Unfortunately not so much for me. Latest fixation was to take the old favourite control problem of balancing an inverted pendulum and using it as a training target for a tiny LLM where state and action are forced through a small fixed vocabulary. After an alarming amount of stuffing around it turns out you can make this work (LQR + swing up heuristic as a teacher and PPO for RL to finesse).
Why you may ask?
to quote Zappa "Without deviation from the norm, progress is not possible." That's my excuse anyway.
More on that later...
Have a happy Friday!
