How To Scale The Use Of Large Language Models In Marketing
Generative artificial intelligence and big language models are poised to transform the marketing industry as we know it.
"To stay competitive, you need to understand technology and how it affects our marketing efforts," Christopher Penn, chief data scientist at TrustInsights.ai, said at the MarTech conference.
Learn how to scale using big language patterns, the value of agile engineering, and how marketers can prepare for the future.
The basis of major language models
Since its inception, ChatGPT has been a popular topic in most industries. You can't go online without seeing everyone's point of view. Few people still understand the underlying technology, Penn said.
ChatGPT is an artificial intelligence chatbot based on OpenAI's GPT-3.5 and GPT-4 language base models (LLM).
LLMs were founded on a premise put forward by the English linguist John Rupert Firth in 1957:
- "You will hear a word of the company he keeps."
This means that the meaning of a word can usually be understood from the words that stand next to it. Simply put, words are defined not only by their vocabulary, but also by the context in which they are used.
This premise is important for understanding natural language processing.
For example, look at the following sentences:
- "I'm making tea."
- - I spilled the tea.
The former refers to a hot drink and the latter is slang for gossip. In these cases, "tea" has many different meanings.
Word order is also important.
- "I'm making tea."
- "The Tea I Made".
Although the above sentences use the same verb "to create", they have different subjects.
How big language patterns work
Below is a diagram of a transformer system, an example of the architecture on which large language models are built.
Simply put, a transformer takes an input and transforms it (ie "transforms") into something else.
It can be used to create LLM, but it is more effective to convert one thing to another.
OpenAI and other software providers provide access to millions of documents, academic articles, news articles, product reviews, forum comments and more. It begins with the collection of a large amount of data, including
Think how many times the phrase "I'll make tea" appears in all these recorded lyrics.
The above Amazon product reviews and Reddit comments are some examples.
Notice the "company" this sentence keeps - that is, any words that appear next to "I'll make tea."
"Taste", "smell", "coffee", "perfume" etc. give context to these LLMs.
Machines cannot read. To process all the text, they use embedding, which is the first step in the Transformer architecture.
Embedding allows patterns to assign a numerical value to each word, and this numerical value appears repeatedly in the text corpus.
Word position is also important for these examples.
In the example above, the numerical values remain the same, but in a different order. This is the postal code.
Simply put, great language patterns work like this:
- Machines record textual data.
- Assign numeric values to all words.
- View statistical frequencies and distributions between different words.
- Try to guess what the next word in the sequence will be.
All of this requires significant computing power, time and resources.
Get a daily newsletter, it relies on search engine marketing.
Agile Engineering: An Indispensable Skill
The more context and guidance we provide to LLMs, the better they are likely to perform. This is the value of agile engineering.
Penn sees the guidelines as safeguards for what machines produce. Machines take the words in our input and combine them as context while expanding the output.
For example, if you are writing ChatGPT instructions, you will find that detailed instructions tend to produce more satisfactory responses.
Writer prompts are kind of like a creative brief. If you want your project done right, don't give your writer one line of instructions.
Instead, send a reasonably sized folder that covers everything you want to write and how you want to write it.
Expand the use of LLMs
When you think of AI chatbots, you might immediately think of a web interface where users can submit requests and wait for the tool to respond. This is what everyone is used to seeing.
“This is by no means the final result of these tools. This is a playground. That's where people can work with the tool," Penn said. "Companies will bring it to market."
Think of typing speed as programming. You are a developer who writes instructions for the computer to do something.
Once you've refined your instructions for specific use cases, you can use the API and have real developers compile those instructions into additional code so you can send and receive data at scale. Programming.
In this way, LLMs will flourish and companies will change for the better.
Since these tools are used everywhere, it's important to remember that everyone is a developer.
This technology will be in Microsoft Office (Word, Excel and PowerPoint) and in many other tools and services we use every day.
"Because programming is done in natural language, the best ideas don't always come from traditional programmers," Penn added.
Because LLMs are based on writing, marketers or PR professionals, rather than programmers, can find innovative ways to use the tools.
How LLMs Affect Search Engine Marketing and What You Can Do About It
We are starting to see the impact of large language models in marketing, especially in search.
In February, Microsoft introduced the new Bing powered by ChatGPT. Users can chat with the search engine and get direct answers to their questions without clicking links.
"You can expect these tools to make your unbranded search much easier because they answer questions in a way that doesn't require a click," Penn said.
"As SEOs, we've seen this before, with snippets and search results without clicks...but it will be even worse for us."
He recommends going to Bing Webmaster Tools or Google Search Console and looking at the percentage of traffic your site is getting from non-brand data searches, as this is the biggest area of SEO risk.
Build your brand
"If branding isn't a top strategic priority for 2023 and beyond, it should be," Penn said.
You need to build your brand and get people to search for your name when they search.
When users are looking for ideas or recommendations on a topic, LLMs will guide them to the summarized information, not you.
But when people specifically search for your name, they're still there.
Make your brand's online presence as strong as possible.
Use AI immune publishing platform.
Penn also emphasized the importance of using a platform that gives you direct and immediate access to your audience.
By using channels like email or SMS (or even direct mail), you can communicate directly with customers and make sure you get there without going through artificial intelligence.
Organic search and social media are already powered by artificial intelligence. Therefore, the probability of reliably reaching even a fraction of your target group is low.
Even the biggest brands can get enough views if they invest in paid campaigns.
Services like Slack, Telegram and Discord allow you to connect with like-minded people and create meaningful connections.
When you add value to your users, you can reliably reach them, earn their loyalty, and increase your brand equity.
Check out: The Marketing Singularity: Great Language Examples and the end of marketing as you know it
At the Penn MarTech conference, he spoke more about the impact of LLMs on marketing jobs. Watch his full presentation here:
Comments
Post a Comment