skeptrune

they're usually pretty bad. a lot of domain specific information just isn't public

I have a backlog of 20 or so to review right now and am getting that out as fast as I can.

Fair

based

the more awesome thing to me is that you can run the MRI through an ensemble of LLMs and check to see if they converge among each other

Well put

This is fun

I really appreciate how "Jony Ive" this looks. Feels like they absolutely nailed the style.

I personally feel like it looks like a disposable tech hardware product, but to each their own. I'm sure a lot of people will love it.

I really thought the photos were real as i was reading. Wow

What a win it is for open source that qwen and kimi show up on this at all.

lmao, i love this

this is awesome. beyond happy to see it

Working on publishing those, but publishing benchmarks requires a lot of attention to detail so it will likely be a bit longer.

agreed. hopefully we can get there soon

Yea we did and actually use Daytona for another product, but it would have been too slow here.

yea chromadb is not the point. multiple data storage solutions work

agreed!

We would also be super interested to see that comparison. I agree that there isn't a specific reason why Chroma would be required to build something like this.

I agree that would have been the way to go given more time and resources. However, setting up a FUSE mount would have taken significantly longer and required additional infrastructure.

100% agree. However, if there were no resource tradeoffs, then a FUSE mount would probably be the way to go.

Modern OCR tooling is quite good. If the knowledge you are adding into your search database is able to be OCR'd then I think the approach we took here is able to be generalized.

Hmmm, the post is an attempt to explain that Mintlify migrated from embedding-retrieval->reranker->LLM to an agent loop with access to call POSIX tools as it desires. Perhaps we didn't provide enough detail?

Vector search has moved from a "complete solution" to just one tool among many which you should likely provide to an agent.

I think it's cool that LLMs can effectively do this kind of categorization on the fly at relatively large scale. When you give the LLM tools beyond just "search", it really is effectively cheating.

100% agree a FUSE mount would be the way to go given more time and resources.

Putting Chroma behind a FUSE adapter was my initial thought when I was implementing this but it was way too slow.

I think we would also need to optimize grep even if we had a FUSE mount.

This was easier in our case, because we didn’t need a 100% POSIX compatibility for our read only docs use case because the agent used only a subset of bash commands anyway to traverse the docs. This also avoids any extra infra overhead or maintenance of EC2 nodes/sandboxes that the agent would have to use.

This is awesome! I'm happy someone made this exist.

Part of why I originally picked programming as a profession is because it felt like a way to get paid to be more of an artisan craftsman. Ultimately, I don't think it was ever quite that, even before ai.

But now post LLM coding agents, its not at all that. Nothing about programming for money resembles artisanship.

It might be time to try sewing wallets or something...

Most useful website I've been on in a hot minute

input: my boss is making me go through a humiliation ritual known as "sprint planning"

output: Grateful for the opportunity to lean into another high-impact sprint planning session today! It’s all about radical transparency, aligning our vision, and embracing the grind to deliver massive value. Growth happens outside the comfort zone! #AgileMindset #Leadership #Synergy

i absolutely never ever in a million years would have guessed this to be the case