Hi, I’ve been thinking about a claim-tracking question, and I’m not sure whether LessWrong has discussed something like this before.
Let’s say someone made a public claim in 2023, and then new evidence in 2025 or 2026 changed the picture quite a bit. How should we label the earlier claim now? Would “outdated” be better, or “partially supported”?
To me, these two are not the same. “Partially supported” sounds more like the claim is still true in some important sense, but some parts are still uncertain. “Outdated” sounds more like the claim may have made sense...
To make the question more concrete, here’s the example I had in mind.
In July 2023, Anthropic's CEO Dario Amodei described AI-bio risk as “emerging and theoretical.” By January 2026, he wrote that mid-2025 internal evaluations showed Claude Opus 4 giving 2–3x uplift on bioweapon-relevant tasks, enough to trigger Anthropic’s ASL-3 threshold.
So I’m curious: would the earlier claim now be better labeled outdated, or partially supported?
My own lean is outdated. It may have made sense in 2023, but it no longer feels like the best description of the situation.
That said, I can also see the argument for partially supported, since the earlier claim did not say the risk was absent.