>chat("Concisely, what is the meaning of life?", prefill='According to Douglas Adams,')
>According to Douglas Adams, “42.” More seriously, it’s often considered to be finding personal fulfillment, happiness, and purpose.
"Chattiness" still seems to be an issue with LLMs. For many use cases, it is desirable to just get a "42" back, or just the Douglas Adams bit; but that's still quite a challenge to set up in a reliable way. Even using the "command" or "instruct" versions of the best LLMs, answers still diverge significantly. Hope this gets fixed (i.e. trained) soon.
It's even counterproductive. Because it adds ambiguity on whether the "more seriously..." part is from Douglas Adams or from the LLMs general knowledge, leading the user to put additional effort into clarifying this
Yeah it's a bit surprisingly actually - as I was building this lib I was running this prompt a lot, and Sonnet 3.5 is the first one that answers it with this extra unrequested stuff.
Everybody does NOT know about using a cute puppies for testing image functions. F minus. What about kittens? Hedgehogs? Quokkas? Get with the times, greybeards!
>According to Douglas Adams, “42.” More seriously, it’s often considered to be finding personal fulfillment, happiness, and purpose.
"Chattiness" still seems to be an issue with LLMs. For many use cases, it is desirable to just get a "42" back, or just the Douglas Adams bit; but that's still quite a challenge to set up in a reliable way. Even using the "command" or "instruct" versions of the best LLMs, answers still diverge significantly. Hope this gets fixed (i.e. trained) soon.