Google's AI Overviews Will Always Be Broken. That's How AI Works

Posted on:
Key Points

A week after its algorithms advised people to eat rocks and put glue on pizza, Google admitted Thursday that it needed to make adjustments to its bold new generative AI search feature..

Googles AI Overviews feature draws on Gemini, a large language model like the one behind OpenAIs ChatGPT, to generate written answers to some search queries by summarizing information found online..

You can get a quick snappy prototype now fairly quickly with an LLM, but to actually make it so that it doesn't tell you to eat rocks takes a lot of work, says Richard Socher, who made key contributions to AI for language as a researcher and, in late 2021, launched an AI-centric search engine called You.com...

In some cases it is better to actually not just give you an answer, or to show you multiple different viewpoints, he says.. Googles head of search Liz Reid said in the companys blog post late Thursday that it did extensive testing ahead of launching AI Overviews..

You.com routinely avoids the kinds of errors displayed by Googles AI Overviews, Socher says, because his company developed about a dozen tricks to keep LLMs from misbehaving when used for search...