The Google search engine powered by generative AI has been activated in the United States to incorrectly answer certain queries that have gone viral. Google a few hours ago explained the reasons why AI Overviews answered some questions about Google Search.
AI Overviews was introduced at I/O 2024 as a piece of text powered primarily by generative artificial intelligence to respond to the user when they perform a search. Nail new experience to adapt to the paradigm shift created by ChatGPT when you respond to queries in a natural conversation with generative AI; and for many users, it’s nicer than having to search for it on the first or second page of Google search results.
In fact, in the explanation given, it clarifies how it works AI Overviews to differentiate yourself in certain key aspects compared to chatbots and LLM products or larger language models.
Google’s AI preview suggests adding glue to make the cheese stick to the pizza, and it turns out the source is an 11-year-old Reddit comment from user F*cksmith 🙂 pic.twitter.com/uDPAbsAKeO
– Peter Yang (@petergyang) May 23, 2024
AI Overviews does not simply generate a response based on the training data, and although it is powered by a custom language model, The model integrates with major web positioning systems which are designed to handle traditional “search” tasks.
Therefore, AI Overviews not only generates text in response to the query, but also include relevant links so the user can further research in the subject. AI previews don’t tend to “panic” (one of the worst problems generative AI chatbots face) or invent things like other LLM products would.
That AI Overviews fails in its response This is caused by three common problems:
- You misunderstand the query.
- By misunderstanding a nuance of language on the web.
- You cannot access valuable information on the subject.
Among all the incorrect answers given last week by the search engine, there was one example that Google used to show what happened: “How many stones should I eat.” Google says it very clearly, does not handle satirical content well.
How are you going to improve it?
If someone asks a question in the search engine, AI Overview takes care of creating a link to an article that In this case, it was just one of the Onions, a satirical medium very popular in the United States. Another response related to this case was the use of glue to attach the cheese to the pizza in order to post the link to Reddit as a first-hand source.
Google also announced the next steps it will take to limit the inclusion of satirical and humorous content such as part of improved mechanisms for detecting meaningless queries. These are three aspects that you will also develop:
- Systems have been updated limit the use of user-generated content in responses that might offer inappropriate advice.
- Activation restrictions added against queries where the AI insights did not offer a useful answer.
- In the news and health topics, new security measures are included. The goal will be to not show AI summaries of important news stories where the facts are real. Additional health restrictions will be included.
Finally, according to 9to5GoogleGoogle has maintained since its publication that the vast majority of AI previews offer high-quality information similar to query answers with the search engine’s well-known rich snippets for years.
