Long-Horizon Consistency
The ability of a model to keep its goals, state, and logic from drifting across dozens of decision steps. It differs from context window size and depends on the model's capacity to compress, update, and verify intermediate states.
Timeline (last 90 days)
In tasks that require maintaining long-horizon consistency, the model exhibits "drift" after being extended to dozens of steps, that is, the goal, state, and logic are not maintained.
The cause of long-horizon consistency failure is not that the context window is not large enough, but that the model's ability to verify its own intermediate states is insufficient, leading to error accumulation.
Long-horizon consistency refers to the model's ability to keep goals, state, and logic from drifting after spanning dozens of decision steps, which is not the same thing as context window size.
All Facts (3)
In tasks that require maintaining long-horizon consistency, the model exhibits "drift" after being extended to dozens of steps, that is, the goal, state, and logic are not maintained.
50%UnverifiedThe cause of long-horizon consistency failure is not that the context window is not large enough, but that the model's ability to verify its own intermediate states is insufficient, leading to error accumulation.
50%UnverifiedLong-horizon consistency refers to the model's ability to keep goals, state, and logic from drifting after spanning dozens of decision steps, which is not the same thing as context window size.
50%