From Newsgroup: comp.misc

Researchers at OpenAI have come to the conclusion, after a careful
mathematical analysis of the nature of those rCLlarge-language modelsrCY
that are all the rage nowadays, that the risk of hallucinations is an unavoidable fundamental characteristic of those models <https://www.computerworld.com/article/4059383/openai-admits-ai-hallucinations-are-mathematically-inevitable-not-just-engineering-flaws.html>:

The researchers demonstrated that hallucinations stemmed from
statistical properties of language model training rather than
implementation flaws. The study established that rCLthe generative
error rate is at least twice the IIV misclassification rate,rCY
where IIV referred to rCLIs-It-ValidrCY and demonstrated mathematical
lower bounds that prove AI systems will always make a certain
percentage of mistakes, no matter how much the technology
improves.

Examples of these problems can be quite embarrassing:

The researchers demonstrated their findings using state-of-the-art
models, including those from OpenAIrCOs competitors. When asked rCLHow
many Ds are in DEEPSEEK?rCY the DeepSeek-V3 model with 600 billion
parameters rCLreturned rCy2rCO or rCy3rCO in ten independent trialsrCY while
Meta AI and Claude 3.7 Sonnet performed similarly, rCLincluding
answers as large as rCy6rCO and rCy7.rCOrCY

I canrCOt believe they were serious about this, though:

rCLUnlike human intelligence, it lacks the humility to acknowledge
uncertainty,rCY said Neil Shah, VP for research and partner at
Counterpoint Technologies.

As we all know, there are *plenty* of humans who lack such humility!
ThatrCOs where concepts like rCLideologyrCY and rCLreligionrCY come in ...
--- Synchronet 3.21a-Linux NewsLink 1.2