Years ago Nicholas Carr argued that Google was making us stupid, that ease of access to information was shortening our attention spans and generally making it hard for us to do “deep reading.” Others worried that search engines were siphoning away readership from newspapers, collecting the cash that otherwise would fund journalism.
Today we’re seeing something similar in software development with large language models (LLMs) like ChatGPT. Developers turn to LLM-driven coding assistants for code completion, answers on how to do things, and more. Along the way, concerns are being raised that LLMs suck training data from sources such as Stack Overflow then divert business away from them, even as developers cede critical thinking to LLMs. Are LLMs making us stupid?
Who trains the trainers?
Peter Nixey, founder of Intentional.io and a top 2% contributor to Stack Overflow, calls out an existential question plaguing LLMs: “What happens when we stop pooling our knowledge with each other and instead pour it straight into The Machine?” By “The Machine,” he’s referring to LLMs, and by “pooling our knowledge” he’s referring to forums like Stack Overflow where developers ask and answer technical questions. ChatGPT and other LLMs have become “smart” by sucking in all that information from sites like Stack Overflow, but that source is quickly drying up.
