@ErsatzCulture

An ironic observation:

The paper gives one list ('SciFi-Constitution-128-AutoMerge') of 106 directives to be used to yield ethical outputs from LLMs.

Giving a robot a triple-digit-length list of ethical imperatives was a plot point in RoboCop 2.

The LLM was only, the authors claim, given the first #RoboCop movie as input.

The idea that there's a secret 107th directive about #Google and #PrincetonUniversity is thus spawned.

#GitHub #USLaw #LLMs #Microsoft #SciFi

@peachfront @cstross

According to the paper:

The people did not read the books. They fed the books (and others) to a computer system in order to construct an LLM that spat out potted summaries (in the forms of questions and answers) of specific plot points (involving AI) in the stories.

The paper states that the LLM sometimes went wrong, giving its condensed form of a Terminator plot point as one example.

#GitHub #USLaw #LLMs #Google #PrincetonUniversity #Microsoft #SciFi #SkyNet

@diazona @cstross

I agree. It is the so-called 'Sci-Fi Benchmark' LLM dataset that, as the paper explicitly says (on pages 96 and 97), is based upon two of Charlie Stross's works. And that's being published on GitHub. Thus, Microsoft.

Not legal advice M. Stross, but one avenue to point your solicitor towards exploring.

@peachfront
#GitHub #USLaw #LLMs #Google #PrincetonUniversity #Microsoft