News for hackers. Stuff that matters.
Re:Legal models hallucinate in 1 out of 6 (or more) benchmar... by yarg on Friday May 31, 2024 @07:49PM (#40541448)
Bonus question 2 is the most ridiculous straw man I've seen in a very long time.

The existence of arbitrary string encodings in transcendental numbers proves absolutely nothing about the processing capabilities of adaptive algorithms.

Re:Legal models hallucinate in 1 out of 6 (or more) benchmar... by pennomi on Friday May 31, 2024 @10:59PM (#40542531)
Exactly. Reading digits of pi doesn’t converge toward anything. (And neither do infinite typewriter monkeys.) Any correct value they get is random, and exceedingly rare.

LLMs corral a similar randomness to statistically answer things correctly more often than not.