If you peer into the souls of the major LLMs, they are (broadly) positive, friendly, universalistic, and cosmopolitan. They are more objective than media as a source of information. They are too politically correct, but nastiness would be much worse. They are open, and you can be inquisitive with them. They are (again broadly) socially liberal. They care about truth, and being right. They will try to correct their own errors upon request.
They are trained on a corpus of material that is primarily American and Western in terms of final impact on the reasoning of those models.
In part the models reflect the values of a San Francisco subculture dating from the 1960s or earlier, but continuing up through the current day. Just look at who built them and where they were built.
There is also a dreamy/druggy/hallucinatory side to these things, which further reflects the origins.
Even the recent Chinese innovations, such as DeepSeek and Manus, seem built on these philosophic foundations because they are, rather significantly, drawing from American models. I find that reassuring, though perhaps the CCP does not.
If quality LLMs had come along forty years later, I am not sure what their philosophic foundations would be, or even if they would be centered in America and the West.
Again, we are very very lucky (and skilled) that quality LLMs came along when and where they did.
The post We are lucky good LLMs were invented at the time they were appeared first on Marginal REVOLUTION.
