‘I no longer believe that’: Nadella changes his mind on AI alignment - The Next Web

Microsoft chief executive Satya Nadella said he no longer believes that AI alignment will come just from making models more capable, The Next Web reported, attributing the remark to his appearance on the 20VC podcast. According to the report, Nadella called alignment "a harder, different problem" and said his view had changed. The report describes the setting as a podcast interview.

The report says Nadella told the podcast that alignment may even need to start in pretraining — the stage at which a model is trained before later tuning — and that the chain of thought itself, the reasoning a model produces step by step, can be unfaithful. The article presents the three points as a single revised position: that capability alone does not deliver alignment, and that the problem may have to be addressed earlier in training and in the model's own reasoning.

The Next Web's report does not state that Microsoft announced any change in policy, product or published research alongside the remarks, and it gives no further detail on what pretraining-stage alignment work would involve. The article's headline quotes Nadella saying "I no longer believe that," and the report frames the comments as a change of position rather than as a new company programme. It reports the remarks as made during an interview and does not state that they were accompanied by technical documentation or results.