Person
James Tillman
Commentary by James Tillman
From the commentary rail — every link leaves the site for the original piece.
- 13 August 2026 · ForethoughtNotes on Implications of Scale-Dependent Algorithms
- 13 July 2026 · ForethoughtNotes on Inference IntegrityClaude Fable’s deliberately triggered sandbagging shows that training-time targets are, by themselves, insufficient to guarantee particular LLM behaviors.
- 4 June 2026 · ForethoughtWhat should go in a model spec?Suppose an AI company is considering whether to include some particular quality X – a rule, virtue, heuristic, default, attitude, goal, or style – in a model spec.
- 13 May 2026 · ForethoughtStickiness in AI Behavioral DesignCurrent model specs aim to shape the behaviors of near-present models. But what if current model behaviors transfer into future models by default?