[treetalker]

> Ironically the one situation I’ve found where Claude noticeably hallucinates is when I ask questions about a large and novel text that’s entirely in context. Which is the opposite of what I would expect to see. …

Sigh.

Accurate responses in such situations would be useful for busy professionals in low-stakes scenarios.

But LLMs cannot replace the effect on the human mind that results from actually reading, understanding, and thinking about a text. There is no substitute: we must do our own thinking, because it is the work that matters — the journey, not the destination, yields the benefits.

[Kiro]

What warrants a "sigh" here? The author is talking about specific situations where the whole text is in context, which I agree is surprising, but you reply with a generic dismissal of LLMs.

[mewpmewp2]

But you can argue with LLMs which will at least motivate me to understand the topic better and also learn new perspectives.

[latexr]

You can also argue with humans, who can motivate you to understand topics better and learn new perspectives. That has been happening since the dawn of humanity.

[dpkirchner]

Possibly, but it takes a while to find those humans with sufficient information (edited to add) and time, sometimes years.

[mewpmewp2]

Yeah, but it's very hard to find someone on demand on that very specific topic that you are very interested in at that given moment. Back and forth takes forever, you have to keep checking your notifications, and some of them resort to personal attacks rather than focusing on the content.