zlacker

This sucks, but it's unlikely to be fixable, given that LLMs don't actually have any comprehension or reasoning capability. Get too far into fine-tuning responses and you're back to "classic" AI problems.