A Metaphor API: A Revolutionary Search Engine Built For LLM

The Internet was a utopia where everyone could access the most up-to-date information on any topic. Still, the fierce competition for users’ attention has distorted the site. The Metaphor team believes this is most evident in the decline of Google search. Ranking higher in Google results to capitalize on the resulting traffic is so important that it has its industry: search engine optimization. As a result, websites compete fiercely not to have the finest content but to rank better in Google’s search results, even for relatively straightforward queries like “eggplant parmesan recipe.”

The Metaphor team set out to restore the enchantment of search by harnessing the power of huge language models; advancements such as GPT3 gave them hope that this was possible. They obtained a startup investment, purchased a GPU cluster, and set out to enhance search. They set out to (and continue to) make it feel like one is being taken by the hand through the sum of all human knowledge when conducting an Internet search.

The group has introduced the Metaphor API, a unified interface for integrating LLM with the web. One can use just a few lines of code:

• Try a keyword or metaphor search
• Parsed HTML will be returned instantly. It is unnecessary to scrape the web.

When doing a metaphor search, a transformer-based model is used to forecast the links that will be most relevant to the query. The main distinction is that the returned results are considerably more tailored to the user’s specific inquiry in Metaphor. If one types “AI podcasts” into Google, they will get a bunch of links like “The 11 Best AI Podcasts,” but in Metaphor, they will get actual podcasts that have been neurally organized by quality and relevancy.

The team’s neural network has been trained to recognize such text and forecast the subsequent connection. The result is a new approach to finding what one needs online that mimics the action of sharing the link one finds. Although it may be initially obscure, searches conducted in this manner can produce relevant and useful outcomes. Here are a few search options:

• Describe or feel the way through a search.
• Only entities of the desired kind will be searched for.
• Discover material that Google doesn’t prominently display because keywords aren’t the best approach or because the search engine needs to give a hoot about ranking it highly.
• Look for more links similar to the one in the search.

Key Features

• For its link prediction capabilities, Metaphor employs a transformer-based architecture. This enables searches to be conducted with the full expressive power of ordinary language.
• It immediately returns rich, parsed HTML for any webpage, so web scraping is never an issue.
• Using the available criteria, one can refine their search by time frame and domain.
• It’s simple to use and comes with Python and Node SDKs. See the implementation guide
• to learn how to let GPT handle everything. 
• Any page in the index can have its content returned instantaneously.
• More results are returned, and LLMs can sort through them.
• Its price is significantly lower than that of the Bing API.
• The first one thousand queries per month using Metaphor are free forever.

Check out the Platform, Reference Article, Tweet, and Try it here. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to join our 28k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.