The use of Large Language Model (LLM) AIs for retrieving factual, objective, information is (rightly) frowned upon, due to their tendency to “hallucinate” in their responses.
The most reliable way to determine objective information, using the internet, is still to use a search engine, to visit multiple links, and then to place varying degrees of trust in what you find.
However, LLMs can be a powerful replacement for search engines when you need to find subjective opinions, thanks to their ability to aggregate multiple data sources.
For example, imagine that you’d like guidance on how to write a wedding speech.
With a search engine you would need to:
- Search for “how to write a wedding speech”.
- Click into the first result.
- Ignore ads and irrelevant content.
- Read, or skim-read, the advice.
- You may disagree with some of the advice, or feel that it’s not relevant, so go to the next result… etc.
With an LLM you can:
- Prompt with “How do I write a wedding speech?”.
- Read a single opinion (response), aggregated from thousands of blog posts, magazine articles, etc.
- If the response is unsuitable in some way, tell the LLM and it will produce a new opinion that is weighted towards your requirements.
LLMs are not going to replace search engines any time soon, but for this search-like use case, they provide a superior experience.
