Any small business owner who spends any time listening to podcasts, reading articles, or perusing blogs about entrepreneurship has likely caught wind of the hype surrounding the recently released GPT3 (Microsoft) and OPT-175B (Meta). Both products are LLMs – Language Learning Machines – designed to mimic human writing for a variety of purposes. Organizations are clamoring to get their chance to play with both products – so much so that there’s a waiting list for GPT3 and its four models (all with different price points depending on sophistication.) Meta has made their OPT-175B available to the public free of charge. There is an application process to get the largest (most sophisticated) package, but other iterations can be obtained through the Huggingface Transformers portal.
It’s no wonder there’s so much demand, given that some are hailing LLMs as “the future of content marketing”, and even replacements for human writers.
The hype is (somewhat) valid. LLMs are capable of doing some pretty impressive things, including imitating the works of popular journalists, writing computer code with extreme accuracy at breakneck speed, and even producing original works of fiction. In addition to these somewhat unexpected functions, LLMs are also capable of serving as chatbots – something most of us have experience with from the customer end. They can copy-edit and serve as a spelling and grammar checker with solid accuracy. They can also translate text between languages with ease given that they are fed documents in multiple languages as part of their training.
Drawbacks for Small Businesses
So small business owners should immediately get on a waitlist or start learning OPT-175B, right?
Well, no. While the idea of never having to write web content, hire a blog writer, or create a product description ever again is enticing but, in this writer’s opinion, GPT3 is nowhere near holding enough value for small business owners and entrepreneurs to focus their time there.
The first limitation to the use of LLMs for small business owners is the cost. Sounds strange, given that one of the best is free, right? Well the term “free” is relative here. Certain basic models are technically open for use free of charge, but they also need extensive training, which means hours of human effort feeding the LLM multiple samples of any type of text it may be asked to replicate. The more material it’s fed, the better it’s going to be, and the closer it will sound to an actual human being. For most small businesses, that’s the goal. If you want your LLM to write blog posts, product descriptions, and web text, you could easily spend dozens of hours training your LLM, and hundreds if you want it to do a wide range of writing, or if you want it to function in multiple languages. If your organization uses a certain type of language and a certain writing tone, as many do in order to establish an organizational culture, your LLM will need to learn this, too. If your LLM sounds generic, customers may catch on quickly to the fact that what they’re reading was not written by a human on your team. At best, they’ll think you’re outsourcing your work. At worst, they’ll realize it is machine-generated text and feel duped and deceived.
Flaws and Restrictions
Even if you spend hundreds of hours training an LLM, there are very real restrictions to what it will be able to produce for you. For one, it’s important to remember that LLMs can’t yet differentiate between what is true and what is false, no matter how many factual pieces of writing you give it to read. This is the biggest ethical concern right now in regards to LLMs. Not only are they capable of repeating stereotypical, misogynistic, homophobic, and otherwise biased language – they are in fact drawn to inflammatory speech in much the same way human beings are. They may make up features when writing product descriptions and can even create radicalized text. They also struggle to present current affairs and up-to-date news information. This leads many scholars to worry that LLMs could easily flood the internet with fake news.
How Corporations May Adopt LLMs
Given the cost of training an LLM, it seems likely that the only adopters in the near future will be researchers hoping to test LLM capabilities and large corporations. Multi-million dollar organizations could possibly use LLMs to create blogs, web content, product descriptions, training manuals, or even articles about their projects. But that doesn’t mean they pose a real threat to small businesses. On the contrary, it’s possible that, as large corporations experiment with LLMs, they will put off customers accidentally.
Most of us have had the experience of engaging with a chatbot. Most of us did not react positively. In general, people don’t like chatbots; they would much prefer to talk to a human being. When a chatbot does not present itself right away and the customer finds out midway through the conversation that they are not talking to a real person, their responses become even more negative because they feel tricked. It seems natural to assume that people will have the same response to bot-generated articles, blog posts, and web content. We like to feel like we’re interacting with another human being, and most people would find it upsetting to realize that the blog they’ve been following for months is written by a machine. In addition, customers may not like the idea of their favorite companies replacing humans with computer programs. Some may decide to take their business elsewhere, seeking out small businesses where they can easily interact with a person.
How LLMs Experimentation Could Benefit Small Businesses
LLMs have flaws. They will make mistakes. They will at times be exposed as machines. As this happens, customers are likely to gravitate away from the large corporations that can afford LLMs and towards small businesses. People are already moving away from big box companies and towards their local shops, and LLMs could push them even further. As customers seek out human connection when shopping, small businesses need to be ready and waiting with smiling team members, unique products, and a social media presence that uses brand-specific language and shows a clear organizational culture. After all, that unique culture is what will keep customers come back. So will engaging, unique, human-generated product descriptions, web text, blog posts, and social media content.
My advice to you? Let corporations play with LLMs. Focus on what makes your company unique, and be ready when customers get fed up with bots and come looking for humans.