Hacker News Reader

Comment by jsemrau | Hacker News Reader

My main observation here is

1. Technically it might be possible to search the Internet, but it might not surface correct and/or useful information.

2. High-value information that would make a research report valuable is rarely public nor free. This holds especially true in capital-intensive or regulated industries.

View on HN

I fully expect one of the AI-related business models going forward to be charging subscriptions for LLM search tool access to those kinds of archives.

ChatGPT plus an extra $30/month for search access to a specific archive would make sense to me.

Kagi is $10/mo. for search and +$15/mo. for premium LLMs with agentic access to search.

What they're talking about is access to professional archives like EBSCOnet or Bloomberg, which usually don't sell to individuals in the first place and start at tens of thousands of dollars per seat for institutional access.

Ofc Mistral model is a lot worse but for $14.99 you get access to AFP news. So OpenAI for $60 for the same things would be a huge joke

The $10 plan includes the LLM assistant a now as well (with a more limited selection of models than the $25 plan).

Then I'd rather see domain-specific agent-first data. I.e., not a simple API call but token->BM25->token

o3/o4 seem to know how to search things like pypi, crates.io, pkg.go.dev etc and apply those changes on the first try. My application (running on an older version of code) had a breaking change to how the event controller functioned in the newer version, o3 looked at the documentation and rewrote it to use the new event controller. It used to be that you were trapped with the LLM being 3-8 months behind on package versions.

Huh, now I'm thinking that maybe a target for release notes should be to provide enough details that a good LLM can be used to apply fixes for any breaking changes.

MCP maybe? A release notes MCP (maybe into ReadTheDocs or pypi) that understands upgrade instructions for every package.

This is the thing I don't really love about MCP: Why should it require a separate protocol, rather than just good readable documentation?

Ironically, https://context7.com/

This is devdocs to be consumed by LLMs, https://github.com/upstash/context7

Brilliant (and one less think I don't have to build)!

I'm glad this exists, but can you describe to me why it needs to? Why can't agents just read the docs directly?

By returning detailed docs for exactly what the AI is coding at the time, it greatly reduces the the likelihood it will make a mistake. It moves from a recall from the training data problem, to a transcription problem.

This is RAG but for API docs.

I'm not a researcher, but don't most researchers these days also upload their work to arXiv?

Sure, it's not a journal - but in some fields (Machine Learning, Math) it seems like everyone also uploads their stuff there. So if the models can crawls sites like arXiv, at least there's some decent stuff to be found.

Not outside of ML, physics, and math. Preprints are extremely rare in many (dare I say most) scientific fields, and of course many times you are interested in not the cutting edge work, but the foundational work in a field from the 60s, 70s, or 80s, all of which is locked behind a paywall. Or at least it's supposed to be, and corporate LLMs are not "allowed" to go poking around on sketchy Russian website for non-paywalled versions.

I'm assuming many AI companies are probably scraping all the PDFs from the "shadow libraries" of the world that have done some of the work of liberating these papers from behind their paywalls. Obviously it's legally unsettled territory right now...

Proper research, especially the one contributed to conferences, is hard to get by and is usually managed by the conference organizers. Arxiv has some, but it's limited.

It would be great if for a DeepSearch tool for ML, I could just use Arxiv as a source and have the Agent search this. But so far I have not found a working Arxiv tool that does this well.

Reply on news.ycombinator.com