We're building a platform for AI friendship. To this end we collected a proprietary dataset of over 4 billion user-bot messages and spent $1 million on compute to train language models optimised for entertainment.