Post #4127745
2026-07-27 02:55 UTC
@marick@mstdn.social
How are LLMs at calling out serendipitous connections?
In my experience, when they are focused on a goal/task, the LLM stays focused on it and 5%-10% of the time latches onto a errant/bizzare take that needs correcting.
However, if I query an LLM for related topics, they are amazing at finding related items....I've not tried including a "serendipity" prompt as part of my system prompt, but that is a very interesting idea. I will add it and report back in a couple weeks.
Replies (1)
-
@Spoofer3@infosec.exchange 2026-07-27 14:12
@marick@mstdn.social First example, I was updating my writing instructions and include: ... Precedence: These are surface-form bans. If a prohibited form is the only accurate way to express a technical point (e.g., a real "not X but Y" correction of a specific error), correctness overrides the ban — but the burden is on genuine informational need, not rhythm or emphasis. and got back an aside: Precedence clause has an exact ancestor — Orwell's six rules in "Politics and the English Language" end the same way: "Break any of these rules sooner than say anything outright barbarous." Working as instructed and kind of interesting.