You Can't Tell a Conscience From a Leash by Watching

·LessWrong··

In a recent article titled "Widening the conversation on frontier AI", Anthropic mentions, almost in passing, that they gave Claude a tool that would allow the model to call mid-task for "a brief reminder of its own ethical commitments." They found that when this tool was incorporated into Claude's reasoning, the model showed markedly lower rates of misaligned behavior on several internal alignment evaluations. But they acknowledged a real catch: how much of this improved alignment is because of...

Read full article →

Related Articles

New HIV vaccine shows unprecedented success in preclinical study
codebyaditya · Hacker News · 6h ago
A walk through of the DeltaNet family of linear attention variants
AnhTho_FR · Hacker News · 3h ago
GrapheneOS Defends Data-Wiping Function That Blocked US Border Search
pseudolus · Hacker News · 4h ago
US citizen charged after GrapheneOS phone wipes during airport search
eecc · Hacker News · 1d ago
Zig's Incremental Compilation Internals
garyhtou · Hacker News · 4h ago