You’ll be able to do the identical thing with words or sentences, as a substitute of images. Notice how within the above example, the vectorization is in a position to capture the semantic representation i.e. it knows that a sentence talking a couple of bird swooping in on a baby chipmunk must be within the (small, animal) quadrant, whereas the sentence talking about yesterday’s storm when a big tree fell on the road must be within the (big, tree) quadrant. In point of fact, there are greater than 2 dimensions — 384 for sentence transformer embeddings (sentences or chunks of text are passed through the BERT transformer, and 384 numbers corresponding to the 384 dimensions are outputted.)

Vector Search + LLM (RAG)

Vector search is now commonplace in all search (Google, Bing, pick your favorite search engine). Take into consideration in case your entire search database has only these 4 documents. You’ll be able to ask the query, “Tell me something about a giant tree.”

The output can be:

During yesterday’s storm, a big tree fell on the road.

Just essentially the most relevant document within the database. But when you asked an LLM the identical query (and append within the 4 “documents” at the top), that is what you get (ChatGPT in this instance):

An enormous tree is usually a big, towering plant with a sturdy trunk, branches, and an intensive network of roots. These trees can reach impressive heights, often extending well above the encircling vegetation. They function essential components of many ecosystems, providing shade, shelter, and food for various organisms.

The incident you mentioned a couple of large tree falling on the road during a storm indicates the ability and impact of severe weather conditions. Storms with strong winds can uproot or break trees, resulting in road blockages and potential hazards for vehicles and pedestrians.

In one other scenario you mentioned, while driving on the road in Bandipur, you encountered a big elephant crossing the road. Bandipur is a national park and wildlife sanctuary positioned in India, known for its wealthy biodiversity and diverse animal population. Elephants are certainly one of the distinguished species present in Bandipur and are known to traverse the world, sometimes crossing roads. Encounters like these highlight the importance of wildlife conservation and the necessity for humans to coexist with animals of their natural habitats.

Switching topics to Bonsai, they’re miniature trees which might be grown in small pots and meticulously shaped and pruned to take care of their small size. Bonsai cultivation is an ancient art form that originated in East Asia and has gained popularity worldwide. Attributable to their limited root space, Bonsai trees require regular watering to stop them from drying out. The small pots could cause the soil to dry quickly, so it’s crucial to watch the moisture levels and water them accordingly.

Finally, you mentioned observing a bird swooping in on a baby chipmunk through your window. This statement reflects a moment of nature in motion, where a bird might need been attempting to find food. Birds of prey and certain bird species, equivalent to hawks or owls, are known to swoop down on small mammals like chipmunks as a part of their natural hunting behavior. It’s fascinating to witness such interactions, as they supply insights into the dynamics of the animal kingdom.

Overall, these snippets of nature — whether or not they involve big trees, elephants, Bonsai trees, or bird predation — highlight the variety and wonder of the natural world that surrounds us.

That answer seemed rather more well rounded. ChatGPT was using its internal knowledge from being trained on trillions of documents, together with the context it was given.

But let’s say you will have greater than just 4 documents — possibly hundreds or hundreds of thousands, you could’t fit into the ChatGPT prompt. On this case, you need to use vector search to narrow down the context probably to contain the reply, append it to the prompt, and ask the identical query as follows:

That is the (truncated) answer it now gives:

You possibly can then have a database, that stores documents and embeddings. You’ll be able to have one other DB that stores queries, and finds essentially the most relevant documents based on the queries:

Once you will have essentially the most similar document(s) by query, you may feed that to any LLM like ChatGPT. By this straightforward trick, you will have augmented your LLM using document retrieval! This can also be referred to as retrieval augmented generation (RAG).