A counter example to this is that I asked it about NovaMin® 5 minutes ago and it essentially told me to not bother and buy whatever toothpaste has >1450 ppm fluoride.

▲

dns_snek 4 days ago | parent | next [-]

Such is the nature of probabilistic systems. Generally speaking, LLMs read the top N search results on the topic in question and uncritically summarize them in their answer. Emphasis on uncritically, therefore the quality of LLM answers is strongly correlated with the quality of top search results.

Relevant blog post: https://housefresh.com/beware-of-the-google-ai-salesman/

▲

simonw 4 days ago | parent [-]

This is why I am so excited about the way GPT-5 uses its search tool.

GPT-4o and most other AI-assisted search systems in the past worked how you describe: they took the top 10 search results and answered uncritically based on those. If the results were junk the answer was too.

GPT-5 Thinking doesn't do that. Take a look at the thinking trace examples I linked to - in many of them it runs a few searches, evaluates the results, finds that they're not credible enough to generate an answer and so continues browsing and searching.

That's why many of the answers take 1-2 minutes to return!

I frequently see it dismiss information from social media and prefer to go to a source with a good reputation for fact-checking (like a credible newspaper) instead.

	▲	Agraillo 4 days ago \| parent [-]
		> finds that they're not credible enough to generate an answer The credibility is one side of the story. In many cases, at least for my curious research, I happen to search for something very niche, so to find at least anything related, an LLM needs to find semantic equivalence between the topic in the query and what the found pages are discussing or explaining. One recent example: in a flat-style web discussion, it may be interesting to somehow visually mark a reply if the comment is from a user who was already in the discussion (at least GP or GGP). I wanted to find some thoughts or talk about this. I had almost no luck with Perplexity, which probably brute-forced dozens of result pages for semantic equivalence comparison, and I also "was not feeling/getting lucky" with Google using keywords, the AROUND operator, and so on. I'm sure there are a couple of blogs and web-technology forums where this was really discussed, but I'm not sure the current indexing technology is semantically aware at scale. It's interesting that sometimes Google is still better, for example, when a topic I’m researching has a couple of specific terms one should be aware of to discuss it seriously. Making them mandatory (with quotes) may produce a small result set to scan with my own eyes.

▲

the_pwner224 4 days ago | parent | prev | next [-]

A year ago I asked it to do deep research on Biomin F + a comparison to NovaMin & fluoride. It gave a comprehensive answer detailing the benefits of BioMin & NovaMin over regular fluroide.

	▲	yeasku 4 days ago \| parent \| next [-]
		A year ago I asked the change of dolar euro and it made up the number. How do you know it did not made it up. Are you an expert in the field?
	▲	typpilol 4 days ago \| parent \| prev [-]
		I'd be curious if you have the same prompt and repeat it what you get.

▲

therein 4 days ago | parent | prev [-]

What's incredible about that is that you are acting like that was a success story but it is a nuanced topic and it swallowed all the nuance and convinced you.

You're now here telling us how it gave you the right answer, which seems to mostly be due to it confirming your bias.