RE: LeoThread 2025-04-23 01:22

Alternatively, could those critics be expecting a transformation as unlikely as turning a duck into a horse simply because enough experts are working on it?

"There's genuine confusion over why some remain so optimistic about large language

models. Daily use of multiple models is intended to save time on searches and summaries, yet they often generate false links, citations, and quotes.

There are instances when a provided quote or reference turns out to be completely nonexistent—links return errors, quotes can't be verified, and even claimed scholarly sources cannot be found.

Although accuracy has improved over the past two years to the point where about 50-60% of references might be authentic, personal observations still favor one model over another, and some models consistently omit references even when

requested. It’s frustrating, especially when even quick magnitude estimates fall short of expectations.

For example, after uploading a paper for summarization, the model incorrectly indicated the paper was from 2023 when it clearly displayed 2025 on the header.

The notion that knowledge graphs could resolve these issues is also misleading, as they don’t address the gap between logical consistency and real-world accuracy.

In time, companies may continue to enhance large language models until a disruptive AI model outperforms them, potentially exposing current overvaluations and severely impacting the market."