SEO is dead! Via… :checks notes: LLMO?

Alright, I'm assuming everyone's been asked at least once ChatGPT question by a client or a stakeholder.

Yes? Hopefully none with budgetary implications (bless their hearts).

Whether your experience was horror or hilarity, it's probably time to unmask the killer catchphrase.

ChatGPT is large learning model (LLM). LLMs are learning algorithms that can recognize, summarize, translate, predict and generate languages using very large text-based datasets.

Remember Furbies? These tiny little deals with the devil traded a toy designer's soul for a set vocabulary, the ability to listen for keywords, and even learn a few new ones. They would respond with a mix of these keywords and "furbish", a translatable pidgin language.

[Fun fact: Mine heard so many jokes about it being possessed, it started saying murder. Truly the perfect childhood toy.]

ChatGPT is a much, much bigger version of 1998's Black Friday trample trophy.

Furby knew 60 parameters ("love","friend","happy", "kill", etc.), ChatGPT knows 175 billion.

But if ChatGPT is posed to pull market share from Google, then that basically leaves SEOs optimizing… for very large furbies.

"Generative AI" takes this new expanded vocabulary and replicates patterns. It's not creative but it's good at moving the words around.

JinaAI does a great job succinctly describing the machinations.

When you enter a query:

1. The LLM processes it with its encoder network, converting the input sequence into a high-dimensional representation.
2. The decoder network then uses this representation, along with its pre-trained weights and attention mechanism, to identify the specific piece of factual information requested by the query and search the LLM's internal representation of this knowledge (or its nearest neighbors).
3. Once the relevant information has been retrieved, the decoder network uses its natural language generation capabilities to compose a response sequence stating this fact.

And that's why I'm pointing to you toward their guide to optimization for large language models.

LLMO is new. This article coins the phrase and offers a rough sketch of mechanics and shared concepts.