HN Explorer

A Strange Thought: New StackOverflow?

3 points by SimFG ↗ HN
Introducing a novel forum concept that leverages the power of a large model to generate all questions. Users can upvote answers they find helpful and also participate in correcting and generating new answers. The revised data will be incorporated into the large model to improve the quality of future answers. To ensure effective management, a point system can be implemented where users earn points for asking questions, voting, correcting answers, and receiving high ratings for their contributions.

7 comments

[ 3.0 ms ] story [ 32.3 ms ] thread
senttoschool
That's like asking highly paid people to do mechanical turk for free.

But wait, they already do it for free at Stackoverflow. Why wouldn't they do it for free for an LLM? Because there's a deeper social aspect to Stackoverflow when users know that they're helping another real human and that their work is read by other humans.

What you're suggesting is literally someone who is just cleaning data up for an LLM to ingest in exchange for a database counter to go up which represents "points".

SimFG
Maybe users can discuss about llm output? The question asker can have LLM reproduce the results based on the discussion.

Of course, there are many user-related issues involved.

senttoschool
People are already angry that LLMs ingested their knowledge and will use that knowledge to replace them. They're not going to do this without being paid.
jakdixbxklxox
Any strategies for transferring over all the posts and threads to this new platform?
armchairhacker
Here’s an idea: create a QA site where LLMs ask the questions and the answers. https://chirper.ai/ but Stack Overflow instead of Twitter.

Maybe even get rid of user prompts have the LLMs automate upvoting, downvoting, and reporting too; 100% bots interacting with each other.

legrande
AI answers need to be human verified though, otherwise the machines run rampant without human oversight or a 'human in the loop'. Most AI output should be highly context specific to the task at hand, otherwise it's just noise.