4 ms·

Until we solve the hallucination problem google search still has a place of power as something that doesn’t hallucinate. And even if we solve this problem of h

Until we solve the hallucination problem google search still has a place of power as something that doesn’t hallucinate. And even if we solve this problem of hallucination, the ai agents still need a platform to do search. If I was Google I’d simply cut off public api access to the search engine.

pixl97 2y ago
>google search still has a place of power as something that doesn’t hallucinate. Google search is fraught with it's own list of problems and crappy results. Acting like it's infallible is certainly an interesting position. >If I was Google I’d simply cut off public api access to the search engine. The convicted monopolist Google? Yea, that will go very well for them.
voidspark 2y ago
LLMs are already grounding their results in Google searches with citations. They have been doing that for a year already. Optional with all the big models from OpenAI, Google, xAI
- asadotzler 2y ago
  And yet they still hallucinate and offer dead links. I've gotten wrong answers to simple historical event and people questions with sources that are entirely fabricated and referencing a dead link to an irrelevant site. Google results don't do that. This is why I use LLM's to help me come up with better searches that I perform and tune myself. That's valuable, the wordsmithing they can do given their solid word and word part statistics.
  - voidspark 2y ago
    Is that using the state of the art reasoning models with Google search enabled? OpenAI o3 Gemini 2.5 Pro Grok 3 Anything below that is obsolete or dumbed down to reduce cost I doubt this feature is actually broken and returning hallucinated links https://ai.google.dev/gemini-api/docs/grounding https://ai.google.dev/gemini-api/docs/grounding
lukeschlather 2y ago
People talk about LLM hallucinations as if they're a new problem, but content mill blog posts existed 15 years ago, and they read like LLM bullshit back then, and they still exist. Clicking through to Google search results typically results in lower-quality information than just asking Gemini 2.5 pro. (which can give you the same links formatted in a more legible fashion if you need to verify.) What people call "AI slop" existed before AI and AI where I control the prompt is getting to be better than what you will find on those sorts of websites.

pixl97 2y ago

>google search still has a place of power as something that doesn’t hallucinate. Google search is fraught with it's own list of problems and crappy results. Acting like it's infallible is certainly an interesting position. >If I was Google I’d simply cut off public api access to the search engine. The convicted monopolist Google? Yea, that will go very well for them.

voidspark 2y ago

LLMs are already grounding their results in Google searches with citations. They have been doing that for a year already. Optional with all the big models from OpenAI, Google, xAI

asadotzler 2y ago
And yet they still hallucinate and offer dead links. I've gotten wrong answers to simple historical event and people questions with sources that are entirely fabricated and referencing a dead link to an irrelevant site. Google results don't do that. This is why I use LLM's to help me come up with better searches that I perform and tune myself. That's valuable, the wordsmithing they can do given their solid word and word part statistics.
- voidspark 2y ago
  Is that using the state of the art reasoning models with Google search enabled? OpenAI o3 Gemini 2.5 Pro Grok 3 Anything below that is obsolete or dumbed down to reduce cost I doubt this feature is actually broken and returning hallucinated links https://ai.google.dev/gemini-api/docs/grounding https://ai.google.dev/gemini-api/docs/grounding

lukeschlather 2y ago

People talk about LLM hallucinations as if they're a new problem, but content mill blog posts existed 15 years ago, and they read like LLM bullshit back then, and they still exist. Clicking through to Google search results typically results in lower-quality information than just asking Gemini 2.5 pro. (which can give you the same links formatted in a more legible fashion if you need to verify.) What people call "AI slop" existed before AI and AI where I control the prompt is getting to be better than what you will find on those sorts of websites.