Hacker News Reader

Comment by tremarley | Hacker News Reader

If you wanted to know more about a new programming language named “Frob” or a plane crash that happened today, couldn’t you use an LLM like grok?

Or any other LLM that’s continuously trained on trending news?

View on HN

How do I know the LLM isn't lying to me? AIs lie all the time, it's impossible for me to trust them. I'd rather just go to the actual source and decide whether to trust it. Odds are pretty good that a programming language's homepage is not lying to me about the language; and I have my trust level for various news sites already calibrated. AIs are garbage-in garbage-out, and a whole boatload of garbage goes into them.

They could provide verbatim snippets surrounded by explanations of relevance.

Instead of the core of the answer coming from the LLM, it could piece together a few relevant contexts and just provide the glue.

They do this already, but the problem is it takes me more time to verify if what they're saying is correct than to just use a search engine. All the LLMs constantly make stuff up & have extremely low precision & recall of information

I don't understand how that's an improvement over a link to a project homepage or a news article. I also don't trust the "verbatim snippet" to actually be verbatim. These things lie a lot.

>Odds are pretty good that a programming language's homepage is not lying to me about the language

Odds are pretty good that, at least for not very popular projects, the homepage's themselves would soon be produced by some LLM, and left at that, warts and all...

> How do I know the LLM isn't lying to me?

How do you know the media isn't lying to you ? It's happened many times before (think pre-war propaganda)

We’re talking about the official website for a programming language, which has to reason to lie.

None of the LLMs (not even Grok) are "continuously trained" on news. A lot of them can run searches for questions that aren't handled by their training data. Here's Grok's page explaining that: https://help.x.com/en/using-x/about-grok

> In responding to user queries, Grok has a unique feature that allows it to decide whether or not to search X public posts and conduct a real-time web search on the Internet. Grok’s access to real-time public X posts allows Grok to respond to user queries with up-to-date information and insights on a wide range of topics.

i can also use my human brain to read a webpage from the source, as the authors intended. not EVERY question on this planet needs to be answered by a high resource intensive LLM. Energy isn’t free you know. :)

Other considerations:

- Visiting the actual website, you’ll see the programming languages logo. That may be a useful memory aide when learning.

- The real website may have diagrams and other things that may not be available in your LLM tool of choice (grok).

- The ACT of browsing to a different web page may help some learners better “compartmentalize” their new knowledge. The human brain works in funny ways.

- i have 0 concerns of a hallucination when readings docs directly from the author/source. Unless they also jumped on the LLM bandwagon lol.

Just because you have a hammer in your hand doesn’t mean you should start trying to hammer everything around you friend. Every tool has its place.

It's just a different kind of data. Even without LLMs, sometimes I want a tutorial, sometimes I want the raw API specification.

For some cases I absolutely prefer an LLM, like discoverability of certain language features or toolkits. But for the details, I'll just google the documentation site (for the new terms that the LLM just taught me about) and then read the actual docs.

Search is best viewed as a black box to transform {user intention} into {desired information}.

I'm hard pressed to construction an argument where, with widely-accessible LLM/LAM technology, that still looks like:

   1. User types in query
   2. Search returns hits
   3. User selects a hit
   4. User looks for information in hit
   5. User has information

Summarization and deep-indexing are too powerful and remove the necessity of steps 2-4.

F.ex. with the API example, why doesn't your future IDE directly surface the API (from its documentation)? Or your future search directly summarize exactly the part of the API spec you need?

I don't know the exact word for this case, but sometimes you want the surrounding information to what you're looking for. Often I skim documentations, books, articles,... not in search for a specific answer but to get the overview of what it discusses. I don't need a summary of a table of contents. But it's a very good tool for quickly locating some specific information. Something like

  Language Implementation Patterns (the book) |> Analyzing Languages (the part) |> Tracking and Identifying Program Symbols (the chapter) |> Resolving Symbols (the section)

  Unit Testing: Principles, Practices,and Patterns (the book) |> Making your tests work for you (the part) |> Mocks and test fragility (the chapter) |> The relationship between mocks and test fragility (the section) |> Intra-system vs. inter-system communications

  Python 3.13.3 Documentation (docs.python.org) |> The Python Standard Library |> Text Processing Services |> string

Could never understand that obsession with summmarization. Sure, it may be be useful for long-form articles or for low-quality content farms, but most of the time you are not reading those.

And if you are reading technical docs, especially good ones, each word is there for a reason. LLM throw some that information away, but they don't have your context to know if the stuff they throw away is useful or not. The text the summary omitted may likely contain an important caveat or detail you really should have known before starting to use that API.

And if you go to a nicely formatted doc page (laravel) or something with diagrams (postgres), it throws all of these away too.

Yes, you can use grok but you could also use a search engine. Their point is that grok would be less convenient than a search engine for the use case of finding Frob's website's homepage.

Perplexity solves this problem perfectly for me. It does the web search, reads the pages, and summarizes the content it found related to my question. Or if it didn't find it, it says that.

I recently configured Chrome to only use google if I prefix my search with a "g ".

Reply on news.ycombinator.com