I asked bard to make a comparison chart of OpenAI, Claude, and PaLM's specs (size and context window) and it repeatedly gave me a nice looking table with obviously wrong values.