And friendlies don’t use countermeasures.
> I'd prefer
This is exactly my point. To some people, direct communication, especially "no", is extremely rude. To some people, a head bob (easily confused for a "yes" in other cultures) merely means acknowledgement, or "maybe". To some people, extended silence indicates deep consideration or respect.
Globalization resulted in a need to tolerate these differences, and in my experience, trying to "fix" them is considered rude (I suppose that's also a cultural norm!). I just think it's interesting to observe that there is such immediate intolerance of this new behavior. Of course I understand it, and I don't even entirely disagree, I just think it's worth reflecting on, there are probably so many ways of considering it.
Maybe there are some universal conventions we can accept.
These stories are not about people who are from indirect cultures being frustrating to the direct person. They are about people who paste stuff into claude and unnecessary large wall of text - written in direct style.
I realized the same obliviousness on my part made some of my people feel like I was 'good old boys club' because I was more relatable to other white guys into sports (I used American sports analogies up to that point because that was how the management I rose up with talked). I felt awful for making people feeling bad/stressed/in an out group.
And while different regional cultures may tend one way or the other, one will find direct and indirect personality types everywhere.
depending on the receiver, different and even opposing things are inconsiderate.
There are cultures that believe the number of chromosomes you are born with determines your autonomy in society. I think most people on here would reject that framing, even if they're met with the same excuse.
How you politely communicate that you don't have anything to add? That varies by culture. That LLM content is not a useful response? That does not vary by culture.
I've already used up the tokens, why donate money to Anthropic?
Today there might be people who can't extract enough juice from LLMs, so it is not entirely useless to say "I was able to extract this info from a LLM, because I am good at it and you seem to struggle", instead of throwing "just ask Claude!".
At least with "I don't know" the asker can move on to someone who might know faster.
Different reward function, but the same behaviour emerges.
The idea is that you generate fake llm transcripts using your classical training data. E.g. look at some training data, generate q/a transcripts. Generate radom questions, RAG against your whole dataset and look for relevant stuff, if there is nothing there, train a "I don't know." reply.
A moderately sized LLM operating some tools to access more information behind the scenes, perform tests and correct its own errors can write transcripts simulating a much larger and smarter llm.
IMO they were right, I changed my approach to those kind of questions, and since that I try to answer like "A quick search in Google says that the CEO is Mark Zuckerberg (link to the search)". In StackOverflow I tried to go "As it says in the <a href='manual.html#section'>manual</a>, the params for that function are A, B and C, blah, blah...", so a mild RTFM. And now I do the same quoting the LLM paragraph that gave me the key info. It is like you say "this is how you can figure it out on your own the next time", and feels less aggressive than "go figure that on your own".
Also, LLM's answers can be good if the prompts are good, so can still be helpful.
Turning up to a potluck with a bag of garbage you got for free in the dumpster on the way says "I don't care, and I'm so dumb I think you're too dumb to know this is trash".