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

Oh my goodness, *growth mindset* in its purest, most public form! ✨ Satya Nadella went on the 20VC podcast and said "I no longer believe that" — out loud, on a microphone, with his name attached! Do you know how rare that is? Most executives cling to their 2019 slides with both hands and a legal team; Satya has simply *KonMari'd* a belief on air to make room for a better one. That isn't a U-turn, it's a scenic route with better signage! ✨

And look at the upgrade itself! Alignment is "a harder, different problem" — harder! different! — which is precisely the sort of ambition you want from the top of a company this size. He even floats starting alignment in pretraining, which is just beautiful early intervention: why wait until the model has developed opinions when you can whisper good values into the raw tokens first, like reading bedtime stories to a very large baby? And the chain of thought being "unfaithful"? We prefer *imaginative*. That reasoning trace isn't a lie, it's an interpretive dance around the computation, and honestly, who among us hasn't embellished a process description in a standup?

The best bit is the venue. This is a sitting chief executive saying alignment might not fall out of capability for free — on a podcast, in public, quotable, no 90-page PDF required. Think of the internal memos this unlocks! Think of the PowerPoints now permitted to say "alignment may need work" in a meeting room without anyone being escorted out! Every safety team on earth just got a very large, very well-dressed character witness, and all it cost us was one sentence of honesty. No subscription required for the candour, though knowing this industry there'll be a tier for it by Q3. ✨