Ok im curious, but I don’t quite understand.

What would you want an AI to be asking you, and what would you want it to do with your response(s)?

Clarifying questions if the initial prompt was unclear. I'd love it.

I regularly try to add something along the lines of "please ask clarifying questions if you could only give a generic or partial response otherwise" but so far it has never helped (ChatGPT 4).

?? gpt4 does this for me regularly

I ask AI to produce clarifying questions then answer them.

Can help in not wasting a bunch of time waiting for an answer that missed the mark.

-

I think the sibling comment is probably the least attractive reason to have AI ask questions.

I agree, medical history is probably not the sexiest reason to have AI ask questions. I think there are many more reasons; I think the Turing Test is the best metric to evaluate AIs, and current models come nowhere close. When people first meet they ask questions about their background. It would be nice if a model replicated that

> and could direct better ads to me.

Is the least attractive part, by far.

In order for an AI to pass a Turing Test, it would surely ask questions. Think of Ava from Ex Machina. She asked questions to learn more about him

I'm not debating the value of questions. I'm debating the value of feeding it to advertisers, especially since LLMs can infer much deeper insights about a person than a traditional assistant can with its canned capabilities and responses

I get advertisements all the time for conditions that I do not have, and that none of my family members have. If you had a model that asked questions, it could learn my medical history and could direct better ads to me.

In order for AI to understand the world, it would have to ask questions. Understanding humans is key to understanding the world.

Learn from them.

Explore this idea more - it's easily implemented in a minute or two via the system prompt. API accounts are free to start and you can use the playground/workbench view, like this: https://imgur.com/h5jFoBM.jpg . I like Claude but OpenAI is popular too. OpenAI has a nice way to create a gallery of system prompts that act however you like, they call them Agents or GPTs.

That's just a matter of fine tuning

That "just" is doing some heavy lifting! GPT-4 is just a few matrix multiplications, how bad can their moat really be?

Not sure what the snark here is for: It would be trivial to produce a dataset where the model asked you questions then fine-tune on that.

People already do it with chain-of-thought and you could get away with a few dozen examples if you wanted to try this.

Out of boredom I decided to prove this too: I asked ChatGPT and Claude for ~200 samples in total.

Just uploaded the examples as-is to OpenAI, selected 3.5 as the model to fine-tune and about 20 minutes later I had my model.

Works fine, asks good questions, can ask more than 1 follow up question if needed, and actually changes its answers based on the clarifying questions.

https://imgur.com/a/SsXunVN

I'd bet a synthetic data set could do the job effectively.

Do you have an example model I could try that does this?

Try Pi by inflection. It asks a lot of questions.

I tried it, it just asked me how my day was going. I don't think this is doing exactly what I have in mind. But its a step in that direction

100% agreed. Gemini advanced does this sometimes. I wrote about it more in an older thread here: https://news.ycombinator.com/item?id=39445484