What does it take to identify tasks in the narrow gap between "trivial" and "impossible" for the world's strongest models?
We're discussing this and more with some of the field's leading researchers at our office in Sunnyvale next week, before we go off to ACL and ICML in July.
Most “hard” problems are useless for training a model.
The useful ones sit in a narrow learnable region where the model fails sometimes and succeeds sometimes.
Here are some early results where we took 5 trivial and 5 impossible cybersecurity environments and had a model
