datavorous
An independent developer who shares AI training experiments on Reddit, exploring code golf and Demoscene-style model fine-tuning.
Timeline (last 90 days)
Developer datavorous shared an experiment on Reddit: by fine-tuning golfed JavaScript code and self-scoring 40,000 samples for RLHF, training a model to explore the syntax of the Demoscene and the possibilities of unfinished code.
datavorous's training is divided into two stages: the first stage is fine-tuning on golfed JavaScript, and the second stage is having the model self-score about 40,000 samples for aesthetics and performing reinforcement learning based on these scores (self-scored RLHF).
Since the original information comes only from a Reddit post title and a brief description, details such as model size, training data volume, and specific evaluation methods cannot be known.
All Facts (3)
Developer datavorous shared an experiment on Reddit: by fine-tuning golfed JavaScript code and self-scoring 40,000 samples for RLHF, training a model to explore the syntax of the Demoscene and the possibilities of unfinished code.
50%Unverifieddatavorous's training is divided into two stages: the first stage is fine-tuning on golfed JavaScript, and the second stage is having the model self-score about 40,000 samples for aesthetics and performing reinforcement learning based on these scores (self-scored RLHF).
50%UnverifiedSince the original information comes only from a Reddit post title and a brief description, details such as model size, training data volume, and specific evaluation methods cannot be known.
50%