When I was a child, Daley Thompson was a sporting legend: a champion decathlete, the generalist of generalists. Good as he was, he couldn’t compete with specialist athletes. In the 100m sprint, for example, his impressive 10.26s personal best was far behind that of his contemporary, the sprinter Carl Lewis.
In software and tools, specialism often outperforms generalists. Specialists optimize for a narrow task, achieving superior performance. However, humans dominate due to our cognitive abilities and tool use — we are “super generalists”.
For language models, will specialists or generalists dominate? I’ve been quite curious about this question for an important reason. Depending on whether specialists win or generalists win, or some combination, will affect the way in which firms like yours will make their investments in generative AI. It will also affect the regulatory frameworks that we might need. Ultimately, it’ll affect the nature of the winners and the shape of the future software industry.
Here is my current thinking.
