LLMs Are Still Toxic, Stuck in the Past, and Bad at Math
Nearly four years after ChatGPT's release, large language models still suffer from the same core flaws — poor math, outdated knowledge, short memory, and toxicity — that were present in GPT-3.5, accor…