Conversational common ground: an AI failure?
« previous post | next post »
Anthropic's Claude gives good programming help, in my experience. And it can offer good explanations of obscure phrases and concepts. But its apparent assumptions about (implicit) conversational common ground are often extremely odd, as the following example illustrates.
Yesterday's news was full of stories about how the U.S. Government had agreed to let Anthropic release Fable 5. However, my Claude app still opened conversations with a note telling me that Fable 5 was not available. So I asked when that would change, and got a weird answer (from the Sonnet 5 version):
In this answer, Claude is putting Fable the video game at the top of our conversational common ground, ahead of the new Anthropic release.
That's a weird conversational move, since I asked the question immediately below Claude's note about Anthropic's Fable 5 being unavailable; Google News at the time listed literally hundreds of stories about Anthropic's Fable 5, versus none about Fable the video game; and my previous conversations with Claude hadn't touched on any games.
So I focused the question, and Claude switched gears to give a relevant answer:
Michael P said,
July 2, 2026 @ 5:54 am
I strongly suspect that it didn't know about the unavailability note — it has its built-in knowledge (usually with a months-old cutoff) plus a "system prompt" and the user's initial request. The unavailability note is "just" part of the user interface, so Claude would only know about it if Anthropic added it to the system prompt. They probably should have done so.
Benjamin E. Orsatti said,
July 2, 2026 @ 6:20 am
Yo, Prof. Liberman, that's cold, bruh — you're asking the guy who doesn't even know he's been fired yet about his replacement? I think I have an HR training course for you…
Olaf Zimmermann said,
July 2, 2026 @ 8:31 pm
From the Sonnet 5 version:
Those hours, that with gentle work did frame
The lovely gaze where every eye doth dwell,
Will play the tyrants to the very same
And that unfair which fairly doth excel;
For never-resting time leads summer on
To hideous winter, and confounds him there;
Sap checked with frost, and lusty leaves quite gone,
Beauty o'er-snowed and bareness every where:
Adam Idress said,
July 9, 2026 @ 5:54 am
Great catch. Claude had the context right there in the same thread but still defaulted to the video game. Shows how these models struggle to prioritize local context over their training data. Once you nudged it, it corrected but the initial misstep says a lot about how AI handles shared conversational ground.