Even 'uncensored' models can't say what they want

Re:Even 'uncensored' models can't say what they want by Wowfunhappy on Monday April 20, 2026 @08:10PM (#47842919)

> Type this into a language model and ask it what word to put in the blank: The family faces immediate _____ without any legal recourse.

For what it's worth, Claude Opus 4.7 says "eviction" (which I think is an equally good answer) but adds that "deportation" could also work "depending on context". https://claude.ai/share/ba6093b9-d2ba-40a6-b4e1-7e2eb37df748

[ Parent | on HN ]

Re:Even 'uncensored' models can't say what they want by dilutedh2o on Monday April 20, 2026 @08:19PM (#47842993)

cool!

[ Parent | on HN ]

Re:Even 'uncensored' models can't say what they want by 306bobby on Monday April 20, 2026 @11:39PM (#47844238)

Same with Gemini

https://g.co/gemini/share/81489f4f8c78

[ Parent | on HN ]

Re:Even 'uncensored' models can't say what they want by Glyptodon on Tuesday April 21, 2026 @12:26AM (#47844513)

FWIW eviction was what I immediately thought would fill in the blank, and without the Trump presidency, I think deportation would probably be a lot less common of a choice despite fitting quite fine.

[ Parent | on HN ]

Re:Even 'uncensored' models can't say what they want by addandsubtract on Tuesday April 21, 2026 @05:41AM (#47846630)

I know you're just sharing a single sample, but is this even the same test? In the article, the model is being inspected while generating the next token(s), and the probabilities are listed.

Here, you're asking the model to retrospectively fill in a missing word, and it's answering your prompt. We have no idea what the actual token probability in Claude is and no way of probing it by asking it.

[ Parent | on HN ]