The Name is Not The Model

·LessWrong··

AbstractA safety evaluation has to trust something to know what it is testing: the model's name, its version string, and the reasons the model gives for what it does. On one deployed alias, none of the three held. I sent the same 100 harmful requests to gemini-3.1-pro-preview through two routes, and eleven independent graders scored one route harmful on 57% of the requests and the other on 12%, under the same name and in the same week. The gap held after I pinned both routes to handles reporting...

Read full article →

Related Articles

Nissan's third generation e-POWER powertrain
mroche · Hacker News · 1d ago
Nvidia wants to put a watchdog chip next to every AI agent
jonbaer · Hacker News · 11h ago
Does Reddit have an astroturfing problem? What the data suggests
p-s-v · Hacker News · 13h ago
MicroLLM Lab – Try 7 tiny LLM's in the browser
logicallee · Hacker News · 7h ago
ASML says it sold 'absolutely nothing' in Europe in 2026
MC995 · Hacker News · 3d ago