Search on Google and you will notice an “AI Overviews” section above the conventional search results. That is the integration of generative AI into search engines, and it is seriously evolving how we experience searching on the internet. While this feature is not universally available yet, it is quite evident that Google aims to embed it into almost every search results page.

The technology behind the AI overviews

The backbone of these AI overviews is the Large Language Model (LLM) engine, a technology which also powers tools like Microsoft’s Copilot, Meta AI, and OpenAI’s ChatGPT. But what exactly is an LLM, and how does it work? Simply put, an LLM is a type of artificial intelligence designed to understand and generate human language.

It works by training on vast amounts of text data, learning the nuances of language, grammar, and context to produce coherent and contextually relevant responses. This training enables LLMs to predict and generate text that closely mimics human writing.

However, what we see in the search engines now is not just LLM in its raw form. It is LLMs integrated with Retrieval-Augmented Generation (RAG) technology. RAG combines the generative capabilities of LLMs with real-time data retrieval from external sources.

In other words, while an LLM can generate text based on its training data, RAG technology enhances this by pulling in up-to-date information from the web. This dual approach ensures that the AI-generated content is not only contextually accurate but also current and relevant.

How these AI overviews are a major leap forward

Google’s implementation of these AI overviews uses this sharp technology to provide users with comprehensive answers to their queries. This ditches the previous way of manually opening up each web result and reading it to get the information you want.

By using an LLM engine integrated with RAG, Google can generate a synthesized response that incorporates the latest information available on the web. This means that when you search for information, the AI overview at the top of the page pulls data from various web pages, cites and credits these sources, and presents a concise and informative summary.

The biggest advantage of this approach is the efficiency it offers to the users. Instead of sifting through multiple web pages to find the information they need, users now receive a well-rounded answer almost instantly. This synthesis of information not only saves time but also provides a more complete understanding of the topic at hand.

Some limitations

However, this AI overviews technology does not come without its limitations. First of all, they can “hallucinate”, which means that they can sometimes generate wrong information even when pulling correct data from reliable web pages. It is a flaw in LLMs in general, and AI overviews suffer from it too.

Therefore, AI overviews should not be relied upon for sensitive matters, especially concerning your life, medical advice, finances, and similar topics.

Secondly, they are particularly effective for informational queries, where users seek knowledge or explanations about a particular subject. For instance, if you're looking to understand the principles of quantum physics or the history of the Roman Empire, AI overviews can provide valuable insights.

On the other hand, if your query involves navigating to a specific brand or purchasing a particular product, AI overviews are less helpful. In such cases, users still need to browse individual web pages to gather detailed information and make informed decisions.

There is a reason they are called "overviews." The term itself suggests a broad summary rather than an in-depth exploration.

Looking forward We are uncertain about how the nature of these "AI overviews" will change in the future. However, one thing is certain: given how extensively AI is being integrated into everything, these overviews are here to stay.

Although they are still in the experimental phase and may continue to evolve, they are not going away. For now, AI overviews provide a glimpse into the future of search, where information is not only at our fingertips but also intelligently curated and presented for maximum utility.