Agree.

>>Both URLs were registered yesterday!<<

The "Leadership" photos all look the same, with the AI-airbrushed style. Also amazing how they both only have three people. What a coincidence.

I can't find any of the supposed "Leadership" team on Linkedin. Some of them have unusual names, which cut through the hundreds / thousands of results.

It's a worrying how easily this sort of BS is becoming. It takes far more time to debunk it then to create it. Now you could basically script this and have 100 pages by this time tomorrow, with different styles and URLs and leadership teams (along with professional portraits)...

If they'd let the URLs age a bit, and maybe not registered them on the same day, used slightly more common names, and changed the styles, it would have been much harder to definitively debunk.

Is it just me or am I seeing more and more of this stuff happening on HN?

Google putting AI search summaries is the worst of all worlds.

You need cheap, fast inference to put on a search results page and finish near the deadline (~3 seconds). Cheap, fast inference is more likely to get stuff wrong. No one is happy.

This has become a common pattern in my Claude & Gemini usage. Always require citations, then check those citations to validate they actually contain the information/data the LLM's output claims. Claude, in particular, seems to make massive logic jumps and trust tertiary data sources way more than it should.

There’s a fun variation in W-Europe that google needs to spend some time on:

Northern Belgium and the Netherlands have web content in the same language. But google uses the content in one lump. Problem is when you search for employment/fiscal/legal/… you constantly get content that applies to the wrong nationality.

This is true of a ton of online discourse. Worse, when the headline of a claim doesn't even match the article it is fronting. I've seen more than a few articles that basically contradict the headline, but end in a "despite all evidence, we think it is correct to say X."

Never believe the citations in Google summaries, I can count on a single hand the time I have found a correct citation.

It's not 2/3 correct, it's <1% correct, in my experience. Maybe it's because I check things that sound off more frequently, but even random checks have not panned out well.

Anybody relying on Google AI summaries is misinforming themselves.

Those summaries (and Bing is at least as culpable) are mostly only good for comedy value. Most favourite genre: returns picture of someone you know alongside biography of completely unrelated person. Least favourite genre: useful stat supposedly cited from a linked reliable source (but turns out to actually be a different stat in the reliable source)

I don’t know if Claude performs similarly from a percentage standpoint, but if you’re using it for search (online or personal docs or wikis), it often also just makes things up.

When you point it out, it’ll do the “ohh you’re absolutely right!” bs. Marketing material and management that believes the material wants to pretend that AI agents are junior employees, but forget that junior employees get fired for doing something like this.

It was true even before AI summaries.

The search result showed a paragraph containing the searched word, the website did not

I did something similar for structured text extraction. I added markers throughout each source document and then, for each piece of info I wanted, I asked the LLM to provide two separate fields:

  xyz
  xyz_citation

The latter was just the node number. So then my code could extract the exact snippet, instead of trusting the LLM to quote something verbatim.

That’s what I did when I built my stuff. I have deterministic content with AI commentary, where it seems most people are doing this crazy thing of sending data through the model. I can’t understand it.

You can just look at nouswise or nblm. They do have such harness behind.

>The intent is good, but it seems they're still in the "fake it until you make it" stage

Please notice the internal contradictions here.

To be honest perplexity does nothing to make sure it's answer are correct let alone the citations. They just look plausible. For anything little bit serious I use nouswise or nblm that sometimes abstain instead of making things up.

> the intent is good

What is that supposed to mean? They're trying to be an llm search engine that's not some radical new concept

Even worse than lawyers is the government doing it: https://newrepublic.com/post/215001/judge-rfk-jr-hhs-fake-ai...

That’s what I feel too. From the references in their PDF, a few links are returning 404s, while others don’t even seem relevant to the so-called “research.” It is pure AI slop. https://hausresearch.com/data/perplexity-citation-audit/perp...

What's surprising about this is that you can get the bullshit machine to produce correct externally validate citations. It's not particularly hard either—it's one of the first things you build when you give an LLM access to a body of documents/search. So for a large public service to whiff like this is certainly a stain on their credibility.