Discussion about this post

User's avatar
Tom Nault's avatar

I don't give a lot of thoughts to the limitations of LLMs. I know that when I can't get what I need, I should try again in a few weeks. It's improving overnight and it happens without much fanfare. It's continually smarter. I never gave much thought to a calculator. I only cared if it could do the calculation and come up with the correct answer and LLMs don't always nail the right answer but they sure get me close. This morning it analyzed my E@RTC blog and it did a masterful job of pattern recognition in my writing and it laid it all out. It was wonderful and it would otherwise take someone months to do the same work it did in five minutes. It pointed me in a new direction and that was delightful.

Harold Fethe's avatar

AI does not "understand" the most powerful intellectual tools in the physical sciences. They're called models of nature.

If you inherited a hundred ounces of pure elemental gold, you could affirm that by sending samples to a dozen analytical labs. Each sample sent as an "unknown" would be a valid experiment. And, each would have 11 others corroborating.

By contrast, two psychiatrists separately examining the same patient have only two chances in three of coming to the same diagnosis. Add a 3rd psychiatrist and the likelihood *drops,* to one chance in three. (Coping With Psychiatric and Psychological Testimony / Ziskin)

Nature yields up its complex reality slowly and unevenly. AI is a text search , connected to a "rules engine." Its methods' contrast with physical science research is stark.

AI will instruct us by overreach. The biggest hedge against AI abuse? The US tort system, which annually awards more damages as a percent of GDP than any other nation.

No posts

Ready for more?