6 Common Mistakes To Avoid When You Start Using AI

If you’re just getting started in AI, you’ll be on a steep learning curve. Understanding the jargon, choosing your favourite LLM, figuring out the tools and automating your workflows isn’t going to happen in a day. Be patient. Understand the terrain before you barrel forward. Taking the time to set the foundations will serve you well in the long run.

I asked AI entrepreneurs and experts to share the mistakes they made when first getting involved, to help you avoid them at the start of your journey. Avoid the pitfalls by reading their wisdom, and take it on board when you reimagine your business with AI.

Mistakes to avoid when you start using AI: experts share

Falling for hallucinations

“ChatGPT writes so confidently and eloquently, it’s tempting to trust the stream of ‘facts’ it provides,” said Dr Jeremy Nguyen, a senior researcher and lecturer in AI from Swinburne University. But exercise caution. “LLMs love to make things up, i.e. ‘hallucinate.’” This is commonplace, so learn to expect it so you’re not caught off guard. “Think of LLMs like an eager intern,” advised Nguyen. “Fact-check their work for anything important. Or even better, ask them to work with source documents you provide, rather than relying on their memory.”

He added that there is a funny side to hallucinations. “In the old days, people thought AIs couldn’t be creative. The hallucination problem is the opposite: they’re too creative!”

Overlooking the human touch

AI entrepreneur and digital creator Anna Poplevina said “While AI is efficient for handling routine inquiries, many customers still seek a personal touch. Relying too much on AI can lead to generic interactions and fail to engage customers who appreciate a human connection.”

But it’s not just customer service. It’s content too. Matt Pogla, founder of AutoGPT, agrees. “You’ve just started using AI to post on social media? We can all tell! Copying and pasting output directly is an AI cardinal sin!” He explained the importance of editing outputs to “include your voice and add the human touch,” and said “Straight AI output is not going to impress anyone, especially when everyone has access to the same tools as you!” A bugbear of Pogla’s is AI-generated comments, which he said are really obvious and don’t add any value.

Ignoring ROI with AI-generated content

Jared Bonilla, chief AI officer at Applied Insights, encourages people diving into AI to think problem-first. He said this means understanding, “what problems do I have that AI can help with?” instead of going straight in with thinking about what AI can do. There’s a big difference.

“Everyone knows AI can create content so many companies start there,” said Bonilla. But when it doesn’t work to increase signups or grow revenue, they call him to help. “While being one of the easiest ways to start using AI, it’s also one of the hardest ways to measure, because there are too many variables to understand success.” Don’t forget about ROI. If you’re using AI tools to produce content, consider the time taken to produce publish-ready pieces, as well as metrics such as views, likes, comments, and conversions, to understand if AI really made a difference.

Handing over private or confidential data

“When email was first introduced, employers trained staff with the guideline ‘if you wouldn’t be comfortable with it being read out to the entire office, don’t send it,’” said Nguyen, who wants you to avoid security pitfalls when using AI. “If you’re not clear on the privacy of the AI model you’re working with, use the same guideline.”

Anything you put into ChatGPT’s interface is passed to OpenAI. This means if you're inputting secret or sensitive material, it's no longer under your control. Nguyen said there are simple solutions. “OpenAI does not train on data passed through the API (use something like TypingMind for that),” as opposed to ChatGPT’s chat interface which will use your data unless you change the settings to opt out. Or Nguyen said you can “work with local models where data never even leaves your machine, such as LLMStudio or ollama.”

Not learning proper prompting

If you’re not happy with the output of an AI tool, it’s almost certainly user error. Lazy prompts bring suboptimal results. Prompting is a skill like anything else in business, but few people learn the rules before getting started. AutoGPT’s Matt Pogla knows, “you’re probably ignoring the power of proper prompting.”

He shared an example of a simple prompt: “write an email about my new service to send for outbound sales reach,” that he said just won’t cut it for great results. “AI tools are not like software we have ever used before. Prompting requires a mastery of language for the best output.” Anna Poplevina added, “AI tools like ChatGPT need customization to match your business's voice and context. Without this step, the LLM might produce content that doesn't resonate with your target audience.”

Upgrading the prompt for much better output involves adding specific elements, demonstrated in Pogla’s new example suggestion: “Take on the role of an expert in cold email outreach. Craft an email about my new service [describe your new service] that creates curiosity and encourages replies. Include variations for different subject lines and CTAs. Keep the language natural and incorporate best practices.”

Forgetting about upkeep

Jared Bonilla has another pitfall for you to avoid. Thinking AI is “one and done.” That’s not the case, he explained. “It’s not as simple as uploading documents to a chatbot and expecting it to be correct 100% of the time in perpetuity.” Every AI tool you use “requires constant maintenance and upkeep.” Underlying technologies evolve, so what once worked perfectly no longer does.

Don’t let your hard work during setup go to waste, don’t underestimate the cost of upkeep in terms of time and cash. Systems will require maintenance. “If you’re going to invest in implementing AI, be prepared to invest in the long run,” adds Bonilla.

Mistakes to avoid when getting started with AI

Avoid wasting time and money by following this guidance, from AI experts who have learned the hard way. Don’t fall for hallucinations, don’t overlook the human touch, measure the return of your investment in AI-generated content and don’t share information you wouldn’t want on a billboard. Learn how to prompt like a pro and consider ongoing costs with every tool you commit to using. Begin your AI journey and scale your business to brand new heights.