Note that the woman on the left is being reported with correctly sexed pronouns while the man on the right is not. The difference? Blog posts. We have seen abominations like "passport sex" and "legal sex," the AP now adds "blog sex" to the list. https://x.com/OsborneInk/status/1891871379240210729
I think it will still take 2-5 years for references to trans to begin to become “suffer from trans” or “trans-delusional”. The deeper problem is that the meaningless phrase “trans woman” or “trans man” is eradicated from language particularly press language because it is propagated into large language model training, and it creates confusion between male and female.
There is no correct version of “trans man” or “trans woman” because it is completely ambiguous whether the referent is actually male or female.
It will make for an interesting requirement which if government employees use, for instance GPT-4o, that all references to “trans man” or “trans female” are adjusted in training data to disambiguate.
[technical: I wrote a paper around 40 years ago about the impact of Usenet and other internet textual sources at the time to “re-encode” human languages by Shannon coding, where terms or entire phrases are substituted by a code, then “reconstituted”. This is exactly what happens in the depths of LLM’s, and exactly how phrases like “trans woman” will distort all LLM generated text over time, merging statistics attributes women don’t have with men and vice-versa.]
Note that the woman on the left is being reported with correctly sexed pronouns while the man on the right is not. The difference? Blog posts. We have seen abominations like "passport sex" and "legal sex," the AP now adds "blog sex" to the list. https://x.com/OsborneInk/status/1891871379240210729
Nice piece, unnerving but well-done.
I think it will still take 2-5 years for references to trans to begin to become “suffer from trans” or “trans-delusional”. The deeper problem is that the meaningless phrase “trans woman” or “trans man” is eradicated from language particularly press language because it is propagated into large language model training, and it creates confusion between male and female.
There is no correct version of “trans man” or “trans woman” because it is completely ambiguous whether the referent is actually male or female.
It will make for an interesting requirement which if government employees use, for instance GPT-4o, that all references to “trans man” or “trans female” are adjusted in training data to disambiguate.
[technical: I wrote a paper around 40 years ago about the impact of Usenet and other internet textual sources at the time to “re-encode” human languages by Shannon coding, where terms or entire phrases are substituted by a code, then “reconstituted”. This is exactly what happens in the depths of LLM’s, and exactly how phrases like “trans woman” will distort all LLM generated text over time, merging statistics attributes women don’t have with men and vice-versa.]