Korva is a scroll of the internet’s best ideas.
AI slop lives in the first output
A tool that needs people
This technology differs from earlier software because it requires people to engage with it, build a relationship with it, and have agency. A technology department can't install it on a desktop and run a workflow through it. It works as a superhuman assistant that helps get thoughts out of a person's head, rationalize them, and present them coherently.
Beating the average
Ask for something and take the first output, and it gives generalities, because these models drive everything toward an average. Getting the best out of them takes a person's knowledge, experience, ideas, and creativity. Jostling with the model and becoming its editor makes the user sharper, and what comes back is amazing.
In investing, creativity is a big part of the craft: taking an obscure or well-organized set of data and finding insights that are not easily observed. AI lets that process iterate in powerful ways.
The isolation worry
Using it every day also raises a worry. The more of the day goes to AI, the less goes to colleagues, and messy human relationships are a rich source of information. Taken to the max, it is isolating.
Why do language models drift toward the average?
These models learn by predicting the most likely next word across enormous amounts of text. When they generate, they tend to favor continuations that are common, so unusual phrasing and specific detail get pulled toward the middle. Researchers have reported that models fine-tuned with human feedback produce less varied output than the base models they started from. The most probable answer to a general question is usually the most generic one.
How does going back and forth change the output?
A language model reads the entire conversation so far as its input each time it responds. Every correction, example, or piece of personal knowledge a user adds becomes part of that input and narrows what the most likely next answer looks like. The first answer is based only on the original request, which is why it is the least specific. The span of text a model can read at once is called its context window.
Why do messy human relationships carry so much information?
Sociologist Mark Granovetter's 1973 paper "The Strength of Weak Ties" found that people often learned about new jobs and ideas through acquaintances rather than close friends. Casual contacts connect to different circles, so they bring in facts that a person's own circle has not already shared. Conversations at work also carry tone, hesitation, and context that never make it into a written document. Researchers in organizational behavior have studied these informal networks for decades.
More from Invest Like the Best with Patrick O'Shaughnessy
