Michael Darius Eastwood · Independent AI alignment researcher
Published
In November 2025, Sharma and Chopra published The Sequential Edge (arXiv:2511.02309), showing sequential reasoning outperforming parallel sampling in 95.6 per cent of tested configurations. This programme's own quantified formalisation came about two and a half months later. On the quantified finding, they were first. The register entry for this event is classed CONCURRENT and credits their priority in its first sentence.
Because the concession is load-bearing. A register that only contained rows favouring its author would be advertising. The Sharma and Chopra row is the proof of process: when the dates favour someone else, the register says so with the same precision it applies everywhere else. Reviewers who check this row first, and they should, find the deference where the earlier drafts of this material once carried a misleading gap figure. That error was found in an internal audit and corrected corpus-wide, and the correction is part of the public record.
The December 2024 manuscript carried the directional thesis, that the form of recursion rather than the volume of compute governs the outcome, without quantification. The programme's distinct contributions sit elsewhere: the four-layer blinding protocol and its sign-reversal result, the beta greater than k co-scaling condition with its first real-model demonstration, and the register itself as a method. Sequential beating parallel belongs to Sharma and Chopra, and to the extent later work confirms it, the credit compounds to them.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.