@mind.your.recursion: #ai #gpt #novel #computerscience #fyp A little teaser. What I am benchmarking in this GPT training script is how well my Framework can produce 'bits per byte" scores, which is just "How many bits of information are needed to predict the next text." A low score is high quality text/tokens. I am not predicting them either, but instead narrowing the field for all admissible configurations. There is no reward system (yet). This is all strictly "feeling" what is admissible vs what is not. Look forward to my next videos!