reasoningfactbearishIn chess-playing models trained with RL, substituting or adding noise to latent thought vectors leaves performance unchanged, while ablating them causes only mild degradationComputation and Language28 Jul 2026http://arxiv.org/abs/2607.20952v1