I was reading an AI system’s thoughts about a potential AI arms race.
The AI system was the Megatron LLB Transformer developed by Nvidia’s Applied Deep Research team.
I agree with its impact analysis. I disagree strongly with its countermeasures.
It says something that I find myself putting together an explanation for myself of how it is wrong.
That’s a big step forward from a system a year or two that made surface-level sense, but where there was no real continuity of thought.
A big scary step.
There are people who are so much brighter than I am that I can’t tell whether or not they are correct. Any accidental flaw in their reasoning I would be unlikely to catch. Any trick I would be unlike to quickly catch.
Should I start worrying about having that same limitation in talking to machines?
It scares me. Scares me a lot.