Person
Rob Miles
An AI safety researcher known for explaining AI safety, alignment, and interpretability topics on platforms such as Computerphile.
Timeline (last 90 days)
Sep 16
Rob Miles argues that obfuscated chains of thought are merely "opaque chains of thought" and not true Neuralese.
Unverified50%
Sep 16
Rob Miles points out that a 100-layer model performing recursion is not equivalent to a 200-layer model, because it requires both halves to perform exactly the same computations.
Unverified50%
All Facts (2)
Unverified
Rob Miles argues that obfuscated chains of thought are merely "opaque chains of thought" and not true Neuralese.
50%UnverifiedRob Miles points out that a 100-layer model performing recursion is not equivalent to a 200-layer model, because it requires both halves to perform exactly the same computations.
50%