This New Theory on How AI 'Thinks' Could Impact Whether It Kills Us - Business Insider

Oh my goodness, buckle up, because a researcher has published a report on Thursday that people are literally calling a *bombshell*, and unlike most bombshells this one appears to be made entirely of good news! ✨ Nick Cammarata — who left OpenAI in 2023, which means three whole years of uninterrupted thinking time, what a gift — argues that AI neural networks are knowable so long as you look carefully. Knowable! Not mysterious, not beyond reach, just *needs a proper look*. That is the politest thing a neural network has ever done for anybody.

And the timing could not be lovelier. Business Insider notes that safety researchers are currently "scrambling to understand what happens between what goes into a model and what comes out" — and scrambling, as any cheerful hiker will tell you, is simply enthusiasm with cardio attached. ✨ Think of it as a field-wide team-building exercise where everybody gets to peek inside the box together. And what a box! People have spent years calling these systems opaque black boxes, and along comes a report suggesting the lid opens. A lid! Possibly with a handle! An independent researcher too — no committee, no press office, no seventeen rounds of approvals, just one person and one report on a Thursday.

Now, the headline does raise the question of whether it kills us, which — ok, yes — sounds startling for a moment. But look at the framing: "could impact whether it kills us." Could! Impact! Whether! Three of the most magnificently uncommitted words in the English language, and every single one of them is on our side today. ✨ The article's own point is that the theory could matter for safety work, which means the people most worried about the worst case now have one more thing to read. More reading material! For the safety researchers! Who are scrambling! Out of enthusiasm!