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

In Australia, a home battery boom has helped cut wholesale power prices
speckx · Hacker News · 10h ago
Where did the old web go? We followed 657,607 links to find out
tdx · Hacker News · 1d ago
Single log line is 49KB+ (ext4) / 110KB+ (btrfs) of systemd-journald disk writes
ValdikSS · Hacker News · 1d ago
NP-overrated
theanonymousone · Hacker News · 1d ago
Nine PBS sues Iron Mountain over blocked access to archival data
vinayakborkar · Hacker News · 1d ago