Singularity AI
Specification, robustness and assurance get collapsed into a single word, and the conflation quietly wrecks otherwise sensible arguments.
"Is the model aligned?" is not one question. It is at least three, and they have different difficulty profiles, different evidence bases, and different people working on them. Collapsing them produces arguments that talk past each other indefinitely.
The specification problem
Can we write down what we want? Mostly not, in closed form. Preference learning exists because direct specification failed — we could not enumerate what a good response looks like, so we learned it from comparisons instead. That works better than hand-written reward functions and inherits a different failure mode: the learned reward model is itself a proxy, with its own exploitable gaps.
The robustness problem
Given a specification, does behaviour hold under distribution shift and adversarial input? This is largely empirical and largely unsolved. Jailbreak research keeps demonstrating that safety training generalises less well than capability training, which is an uncomfortable asymmetry nobody has a principled account of.
The assurance problem
Can we demonstrate the first two to a third party who does not trust us? This is where interpretability, evaluation and audit meet governance, and it is the least mature of the three. It is also the one regulation actually requires.
Why the conflation is costly
Consider a common exchange. One party says alignment is "basically working" — pointing at RLHF-trained models being broadly helpful and rarely harmful in ordinary use. Another says it is "nowhere near solved" — pointing at jailbreaks and the absence of any assurance method worth the name.
Both are correct. They are answering different questions. The first is a claim about specification being tractable enough for deployment; the second is a claim about robustness and assurance being immature. The disagreement is almost entirely definitional, and it consumes a remarkable amount of oxygen.
Ask which of the three someone means before deciding whether they are being naive or alarmist. Usually they are neither.
A modest proposal
Write the word less. "Specification is adequate for this deployment context but we have no assurance story" is longer than "it's aligned", and it is the sort of sentence that lets a conversation make progress.