Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

masonyarbrough.com

Post reply on HN
Print view View original

31–40 of 65 posts

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#31

Post by mgraczyk »

echollama wrote: post #27
mgraczyk wrote: post #10

Earlier quoted context omitted.

But then that means they are editing a markdown file on your computer? How is that meant to work? I like the idea but would rather it use Slack or something if it's meant to ask anyone.

this is mainly meant as a way to conversate with the model while you are programming with it. This is not meant to pull questions to a team but more to pair program. a markdown file is best for syntax in an llm prompt and also just easiest to have open and answer questions with. If i had more time and could i would build an extension into cursor.

Why not have the model ask in the chat? It's a lot easier to just talk to it than open a file. The article mentions cursor so it sounds like you're already using cursor?

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#32

Post by geraneum »

echollama wrote: post #28
threeseed wrote: post #22

> an mcp server that lets the agent raise its hand instead of hallucinating a) It doesn't know when it's hallucinating. b) It can't provide you with any accurate confidence score for any answer. c) Your library is still useful but any claim that you can make solutions more robust is a lie. Probably good enough to get into YC / raise VC though.

reasoning models know when they are close to hallucinating because they are lacking context or understanding and know that they could solve this with a question. this is a streamlined implementation of a interanlly scrapped together tool that i decided to open-source for people to either us or build off of.

> reasoning models know when they are close to hallucinating because they are lacking context or understanding and know that they could solve this with a question.

I’m interested. Where can I read more about this?

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#34

Post by olalonde »

loloquwowndueo wrote: post #3

- someone sets up an “ask human as a service mcp” - demand quickly outstrips offer of humans willing to help bots - someone else hooks up AI to the “ask human saas” - we now have a full loop of machines asking machines

I built this - but mostly as a joke / proof-of-concept: https://github.com/olalonde/mcp-human

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#35

Post by threeseed »

echollama wrote: post #28
threeseed wrote: post #22

> an mcp server that lets the agent raise its hand instead of hallucinating a) It doesn't know when it's hallucinating. b) It can't provide you with any accurate confidence score for any answer. c) Your library is still useful but any claim that you can make solutions more robust is a lie. Probably good enough to get into YC / raise VC though.

reasoning models know when they are close to hallucinating because they are lacking context or understanding and know that they could solve this with a question. this is a streamlined implementation of a interanlly scrapped together tool that i decided to open-source for people to either us or build off of.

> reasoning models know when they are close to hallucinating because they are lacking context or understanding and know that they could solve this with a question

You've just described AGI.

If this were possible you could create an MCP server that has a continually updated list of FAQ of everything that the model doesn't know.

Over time it would learn everything.

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#36

Post by echollama »

mgraczyk wrote: post #31
echollama wrote: post #27

Earlier quoted context omitted.

this is mainly meant as a way to conversate with the model while you are programming with it. This is not meant to pull questions to a team but more to pair program. a markdown file is best for syntax in an llm prompt and also just easiest to have open and answer questions with. If i had more time and could i would build an extension into cursor.

Why not have the model ask in the chat? It's a lot easier to just talk to it than open a file. The article mentions cursor so it sounds like you're already using cursor?

would probably work better, this is just how i threw it together as an internal tool a long time ago. i just improved it and shipped it to opensource it.

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#37

Post by lordmauve »

loloquwowndueo wrote: post #3

- someone sets up an “ask human as a service mcp” - demand quickly outstrips offer of humans willing to help bots - someone else hooks up AI to the “ask human saas” - we now have a full loop of machines asking machines

Finally, the "AI" turns out to be 700 Indians. We now have the full loop of humans asking machines asking humans pretending to be machines. Civilisation collapses

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#38

Post by aziaziazi »

olalonde wrote: post #34
loloquwowndueo wrote: post #3

- someone sets up an “ask human as a service mcp” - demand quickly outstrips offer of humans willing to help bots - someone else hooks up AI to the “ask human saas” - we now have a full loop of machines asking machines

I built this - but mostly as a joke / proof-of-concept: https://github.com/olalonde/mcp-human

Cool project! Naive question: does mechanical turk uses llm now?

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#39

Post by aziaziazi »

kjhughes wrote: post #6

Cool conceptually, but how exactly does the agent know when it's unsure or stuck?

I had the same question reading your post:

> (problem description) your agent […] makes confident assumptions

> (solution description) when it’s unsure

I read this as a contradiction: in one sentence you describe the problem as an agent being confident while hallucinating and in the next phrase the solution is that the agent can ask you if it’s unsure.

You tool is interesting but you may consider rephrasing that part.

Re: Show HN: Ask-human-mcp – zero-config human-in-loop hatch to stop hallucinations

#40

Post by atoav »

I am running an electronics/medialab in an university, the amount of fires bad electronics advice from LLMs caused already is probably non-zero.

It is amazing how bad LLMs are when it comes to reasoning about simple dynamics within trivial electronic circuits and how eager they are to insist the opposite of how things work in the real world is the secured truth.

Post reply on HN