I asked gemini to do a deep research on the role of healthcare insurance companies in the decline of general practicioners in the Netherlands. It based its premise mostly on blogs and whitepapers on company websites, who's job it is to sell automation-software.

AI really needs better source-validation. Not just to combat the hallucination of sources (which gemini seems to do 80% of the time), but also to combat low quality sources that happen to correlate well to the question in the prompt.

It's similar to Google having to fight SEO spam blogs, they now need to do the same in the output of their models.

Better source validation is one of the main reasons I'm excited about GPT-5 Thinking for this. It would be interesting to try your Gemini prompts against that and see how the results compare.

I've found GPT-5 Thinking to perform worse than o3 did in tasks of a similar nature. It makes more bad assumptions that de-rail the train of thought.

I think the key is prompting, and bound boxing assumptions.

Kagi has some tooling for this. You can set web access “lenses” that limit the results to “academic”, “forums”, etc.

Kagi also tells you the percentages “used” for each source and cites them in line.

It’s not perfect, but it’s a lot better to narrow down what you want to get out of your prompt.

When using AI models through Kagi Assistant you can tweak the searches the LLM does with your Kagi settings (search only academic, block bullshit websites and such) which is nice. And I can chose models from many providers.

No API access though so you're stuck talking with it through the webapp.

It would be interesting to see if that same question against GPT-5 Thinking produces notably better results.

But the really tricky thing is, that sometimes it _is_ these kinds of forums where you find the best stuff.

When LLMs really started to show themselves, there was a big debate about what is truth, with even HN joining in on heated debates on the number of sexes or genders a dog may have and if it was okay or not for ChatGPT to respond with a binary answer.

On one hand, I did found those discussions insufferable, but the deeper question - what is truth and how do we automated the extraction of truth from corpora - is super important and somehow completely disappeared from the LLM discourse.

Seems like the right outcome was had, by reviewing sources. I wish it went one step further and loaded those source pages and scroll/highlight the snippets where it pulled information from. That way we can easily double check at least some aspects of it's response, and content+ads can be attributed to the publisher.

In the absence of easily found authoritative information from the manufacturer, this would have been my source of information. Internet banter might actually be the best available information.