Decoder with Nilay Patel · Thursday, September 17, 2026
Mustafa Suleyman argues that the steerability of AI models has improved over the past three years, contrary to concerns about alignment failures. He points to the models' better adherence to complex instructions and reduced instances of hallucinations and bias as evidence.
“if you look back over the last three years, the main change in my opinion that has driven progress is that the models have become more steerable. They follow instructions and you can set more and more complex goals for them that require them to act accurately over multiple timesteps using all sorts of tools.”
“That is evidence that we have got more alignment over the last three years, not less. We don't so much talk about hallucinations or bias or all of these other needles that we had in the previous generations.”