Might I suggest a name for anyone building yet another Large Language Model (LLM)? Mediocrates would be a far more honest name than the ones big tech companies choose.
Why Mediocrates? Who was Mediocrates?
Legends say Mediocrates was Socrates’ younger brother. He was supposed to step in for Socrates when the great philosopher drank the cup of hemlock, but pursuing the truth was much too difficult. Instead, he decided to pursue practical knowledge and do so in the easiest way possible.
Dear ChatGPT, Please Do All My Thinking for Me, Thanks!
Mediocrates decided reading books so he could learn to build solid arguments and elegant writing skills was far too much trouble. It is easier to let a chatbot do your writing for you. Reading an entire book or research paper was for chums. He decided too many books and papers were being published anyway, so he would let a chatbot summarize new books and papers.
So why call a new chatbot, Mediocrates? Because chatbots are driving us toward mediocrity in at least three ways.
First, people learn to write by reading. If you read excellent writing, you will pick up excellent habits and a broad vocabulary and become an excellent writer. LLMs work by taking a weighted sum of all the writing collected from the Internet. The broad spectrum of writing across the Internet will vary from excellent to sheer junk; a weighted average between excellent and junk will be the median — the mediocre.
If people learn to write by reading, and their primary reading is chatbot output, they will imitate the mediocre style of the chatbot, becoming mediocre writers.
Why Bother Reading?
Second, we can extend this to excellent thinking. Chatbots aim to give us practical information right now. Using a chatbot to summarize research and books means we don’t need to do the hard work of reading ourselves. As Mediocrates might say, “Who has time to read a long, hard book? I have a television show to binge on!”
Thinking is a skill. Intellectual virtue is developed across many years of study and work. Just like any other muscle, the brain needs exercise to develop strength and scope. Hard reading helps us develop the mental maps we need to navigate the world of ideas. Consuming information in small, easy-to-read summaries helps us move toward mediocrity.
Finally, chatbots are computers. Computers are like scientists; they accurately and dispassionately search for truth without bias. Or maybe they search for a consensus rather than the truth — but a consensus is good enough, right? The chatbot teaches there is no truth, just a consensus of what experts say. We don’t need to question anything or find out the facts independently because the experts will never steer us wrong.
Many researchers are investigating chatbot bias and thinking about what happens when a chatbot begins to feed on its text — but widespread cultural damage doesn’t even require bias or system collapse. Just driving a broad part of a population towards mediocrity might be damaging enough.