Mainebiz

W W W. M A I N E B I Z . B I Z 21 S E P T E M B E R 7 , 2 0 2 6 1/2 H F O C U S S TA R T U P S / E N T R E P R E N E U R S H I P For the life you choose. 1.888.725.2207 norwaysavings.bank/lets-talk M E M B E R F D I C Consistency that helps move your business forward. Running a business means navigating change. Markets shift. Plans evolve. But strong business relationships should be constant. For 160 years, Norway Savings Bank has supported businesses through every cycle, every challenge, and every opportunity. Our team is built on consistency—local experts who know your business, understand your industry, and are there to help keep it moving forward. Because when everything else changes, the right team stays with you. Plans change. Relationships endure. Business banking that starts with you. Luke Hiebert, Sport Thoma and Jack Day, Norway Savings Commercial Lender How can you get AI to stop making stuff up? I've found a few approaches that can greatly improve the veracity of your favorite LLM's output, and I use them whenever drafting or editing content for our website, blog, or other communication channels. Actions to take Bring your own documentation. Rather than relying on the internet to inform AI, provide it with documents you created or can vouch for. is is called RAG, or Retrieval-Augmented Generation. Upload your documents and instruct AI to work solely off documentation you provide, such as product spec sheets, government regulations and profit and loss statements. Don't guess, don't infer and don't assumae. Just the facts, ma'am. is is probably the No. 1 action you can take to reduce hallucinations. Give AI permission to say, 'I don't know.' Tell your LLM to inform you it's not sure rather than guessing. You can bake this into the prompt or train the model to use this approach. Create 'hand-off docs' for longer chats. AI can only keep so much information in its memory at one time. Very long chats cause AI to drop certain parts of the con- versation. To avoid this, occasionally ask AI to create a hand-off doc that can be pasted into a fresh chat window and pick up the conversation from there. Fact-check in a new chat … or even a differ- ent model. e first time I used AI to do research for a TV segment I was preparing, I was terrified. I had always relied on my own research, and what if the AI output was filled with fallacies? It was my name and my face on TV, not the model's. I took the output and I ran it through three dif- ferent AI models asking to review all the facts pro- vided, research them, and identify them as "truth- ful," "likely truthful but unprovable," and "false." ankfully, all three reported everything as truthful with one item as likely truthful but unprovable. Keep a human in the loop. Even when I've taken every step listed here (and a few others), I don't get 100% accurate output. Because AI is a pre- dictive technology, because of the way it "reads" documents, and because it's a black box that even its creators don't entirely understand, mistakes, hallucinations and outright lies will still occur: "You're right, and that's on me." How many times has your AI said that to you when you pointed out a mis- take? It's your expertise, your real-world experience and your cognitive ability that allow you to catch mistakes that otherwise might end up in print. Better safe than sorry It only takes one hallucination to destroy your repu- tation and become a cautionary tale. By taking pre- cautions, you can tap into the power of AI without sacrificing the credibility your audience expects. Rich Brooks is the founder and president o f f l y t e n e w m e d i a , a d i g i t a l a g e n c y i n P o r t l a n d t h a t h e l p s b u s i n e s s e s w i t h branding, web design and digital marketing. It only takes one hallucination to destroy your reputation and become a cautionary tale.

• Cover