I mean, how many people fact check a book? Even at the most basic level of reading the citations, finding the sources the book cited, and making sure they say what the book claims they say?
In the vast majority of cases, when we read a book, we trust the editors to fact check.
AI has no editors and generates false statements all the time because it has no ability to tell true statements from false. Which is why letting an AI summarize sources, instead of reading those sources for yourself, introduces one very large procedurally generated point of failure.
But let’s not pretend the average person fact checks anything. The average person decides who they trust and relies on their trust in that person or source rather than fact checking themselves.
This is a two part problem. The first is that LLMs are going to give you shoddy results riddled with errors. This is known. Would you pick up a book and take it as the truth if analysis of the author’s work said 50% of their facts are wrong?The second part is that the asker has no intent to verify the LLM’s output, they likely just want the output and be done with it. No critical thinking required. The recipient is only interested in a copy-paste way of transferring info.
If someone takes the time to actually read and process a book with the intent of absorbing and adding to their knowledge, mentally they take the time to balance what they read with what they know and hopefully cross referencing that information internally and gauging it with “that sounds right” at least, but hopefully by reading more.
These are not the same thing. Books and LLMs are not the same. Anyone can read the exact same book and offer a critical analysis. Anyone asking an LLM a question might get an entirely different response depending on minor differences in asking.
Sure, you can copy-paste from a book, but if you haven’t read it, then yeah…that’s like copy-pasting an LLM response. No intent of learning, no critical thought, etc.
I mean, how many people fact check a book? Even at the most basic level of reading the citations, finding the sources the book cited, and making sure they say what the book claims they say?
In the vast majority of cases, when we read a book, we trust the editors to fact check.
AI has no editors and generates false statements all the time because it has no ability to tell true statements from false. Which is why letting an AI summarize sources, instead of reading those sources for yourself, introduces one very large procedurally generated point of failure.
But let’s not pretend the average person fact checks anything. The average person decides who they trust and relies on their trust in that person or source rather than fact checking themselves.
Which is one of the many reasons why Trump won.
This is a two part problem. The first is that LLMs are going to give you shoddy results riddled with errors. This is known. Would you pick up a book and take it as the truth if analysis of the author’s work said 50% of their facts are wrong?The second part is that the asker has no intent to verify the LLM’s output, they likely just want the output and be done with it. No critical thinking required. The recipient is only interested in a copy-paste way of transferring info.
If someone takes the time to actually read and process a book with the intent of absorbing and adding to their knowledge, mentally they take the time to balance what they read with what they know and hopefully cross referencing that information internally and gauging it with “that sounds right” at least, but hopefully by reading more.
These are not the same thing. Books and LLMs are not the same. Anyone can read the exact same book and offer a critical analysis. Anyone asking an LLM a question might get an entirely different response depending on minor differences in asking.
Sure, you can copy-paste from a book, but if you haven’t read it, then yeah…that’s like copy-pasting an LLM response. No intent of learning, no critical thought, etc.