LLM reading

After hundreds or thousands of hours of experiences of using different LLM for text reding, it is safe for me to say that the pattern is smart but “lazy”. The worst thing I noticed is that they tend to mix the person and their activities. For example, “Lisa has a friend Jack. Jack is rich.” After several iterations, LLMs will say something like “Lisa is rich”. It sounds like simply combining those words and terms, without further checking or logical filtering. I meant “A 2 years old daughter drives her mom home” is not only showing how lazy LLMs are but also doesn’t make any sense. I hope this kind of ridiculous story will be less in the future.