Thinking Inside the Box
Hollis Robbins,
Anecdotal Value,
Nov 14, 2026
It would be good to have benchmarks for AI performance, says Hollis Robbins, but "benchmarks are a little like learning outcomes, I am sad to say."They are also very conservative. "The downside (of benchmarks) is that they can organize (and limit) learning around performance recognized in advance. The system can be gamed." Again, just like learning outcomes. "All the hundreds of thousands of American college students in pre-packaged online classes are similarly getting only consensus thinking. These are the next generation of AI users. It could be consensus + consensus + consensus all the way down." That would indeed be sad.
Today: Total: [] [Share]

