Presumably the article is about something. So, go to the canonical site/docs about that thing itself. (Trace the source "upstream".) Find where that community holds its discussions. (Mailing list? Discord? Blogs? Github? Scholarly journals?)

Also, if you think Google search gives surface-level results and is a poor option, I hate to say it, but I highly suggest learning how to use Google better, if that applies. It sounds like this is not you, but IMO Google only sucks badly if you use it wrong? I mean, I get the SEO-enshittification and the censorship problem and all that (or: building for the masses and for advertisers), but for most topics it works well enough if used as a power user (so, not just pasting in what you are looking for), IME. You probably know all this, but:

  - require keywords or key phrases with quotation marks
  - AND your required text: multiple phrases, each about a "concept" from the article, or something that marks/signifies of a related concept or take you want  to know more about
  - use minus sign for negative required words (so, excluded quoted phrases) if needed
  - (Some of these used to not work for a while but I think that was along time ago and they pretty much work again?)
  - restrict the query to specific domains or TLDs (site:youtube.com; or *.edu or *.gov, etc)
  - etc.

I also don't see how discussing the subject with a frontier LLM is bad or surface-level. If the material was trained on, it's often a great method. What exactly makes that a poor option "nowadays"? IME this option is better than ever. (But same as with search: it can help to nudge the LLM, even if ever so little, into being more useful, by giving it a little more to work with that literally; just the original topic itself -- what do you want to know about that topic?)

Can you give a concrete example where the standard methods fail?

Finally, searching HN itself is a great option IMO.