SCORE Nashville hosted a 2-hour generative AI workshop focused on practical GenAI use for small businesses.
Presenter Sandy DeWalt (SCORE subject-matter expert) covered GenAI fundamentals, model types, benefits, limitations, prompting techniques, legal/copyright considerations, and demonstrated ChatGPT.
Emphasis: start with free GenAI tools, use models as assistants (human in the loop), and always refine outputs for business-specific voice and accuracy.
Action Items
(Immediate – All attendees) Try GenAI on the next piece of content you write and practice prompting.
(Immediate – Attendee) Sign up for a SCORE mentor via local chapter website if you don’t have one.
(Near term – Attendee) Save effective prompts (in Google Doc/Word) for reuse and refinement.
(When needed – Attendee) If using paid GenAI features (plugins, longer context), evaluate vendor privacy/data controls before uploading sensitive data.
Workshop Agenda / Topics
SCORE Overview
SCORE: national volunteer nonprofit; partner to SBA; provides mentoring and education.
Mentors = free business coaches; use local chapter assignment for local help.
SCORE offers workshops, webinars, pitch competitions, special events.
Presenter Background
Sandy DeWalt: product development, product marketing, pricing, IP subject-matter expert; teaches GenAI, pricing, product development, and IP workshops.
GenAI Types & Focus
Two main AI types:
Generative (text-based) LLMs: chatGPT, Gemini, Copilot, Claude — focus of workshop.
Visual/image/audio/video models: DALL·E, MidJourney, Stable Diffusion, Adobe Firefly — less emphasis for most small businesses.
Workshop is model-agnostic: fundamentals apply across platforms.
Why Start With Text GenAI
Best ROI for small businesses: drafting, ideation, editing, repurposing content.
Free tiers are often sufficient to begin; upgrade only when necessary.
How GenAI Works (Fundamentals)
Large Language Models (LLMs) are pre-trained on vast language corpora.
Transformer layer enables natural-language prompts from non-coders to interact with models.
Models predict next-word probabilities rather than “thinking”; outputs are probabilistic and variable.
Models have training-data cutoffs; they may not know recent events without user-provided context or plugins.
Action: specific deliverable (blog post, 200 words, outline, checklist), platform, tone, length.
Refine/Reiterate: ask for edits, alternatives, different tones, and repurposing (social posts, summaries).
Persona types useful for small businesses
Expert persona: model acts as domain expert for drafting or advising.
Audience persona: model takes the target customer's perspective to surface assumptions and language.
Prompt tips
Save effective prompts; combine elements in natural language.
Ask model to “take more time to review” when analyzing multiple files or needing thorough output.
Use constraints (word limits, remove emojis) to match brand voice.
Demonstration Highlights (ChatGPT)
Free ChatGPT interface: chat window, upload files (free tier may limit file uploads to one per 24 hours).
You can upload PDFs/Word to ask for summaries, outlines, or Q&A about the content.
Settings include data controls to opt out of allowing your inputs to be used for model training.
Practical pattern demonstrated: provide persona + context + action + ask for shorter versions or rewrites.
Benefits For Small Business
Increases productivity: generate first drafts, outlines, and checklists quickly.
Reduces tedious tasks: templates, surveys, email sequences, and repurposing content.
Acts as an independent editor: grammar, tone, clarity, and content critique.
Supports ideation and problem-solving: brainstorm titles, campaign ideas, alternate phrasings.
Enables repurposing: adapt one piece of content to multiple platforms with speed.
Limitations And Risks
Variation: same prompt can yield different outputs; randomness exists.
Cutoff Dates / Offline Knowledge: most models trained on data up to a cutoff; recent events require user-supplied details or special plugins.
Context Length / Token Limits:
Models have token (context) limits per conversation thread; long documents may exceed capacity.
If context limit exceeded, model may “lose the thread” and hallucinate confidently.
Not a search engine substitute: models don’t reliably provide live sources; use specific research tools (Perplexity, research-grade models) for sourcing.
Long-form generation: models can help structure but aren’t well-suited for generating entire books in one pass.
Biases: models inherit biases from training data (notably on image generation and sometimes text).
Regulatory caution: sectors like healthcare and legal require additional care and compliance guidance.
Legal And Copyright Guidance (Non-Legal Summary)
Human authorship matters: US Copyright Office guidance currently requires human authorship for copyright protection.
Pure AI output (unchanged) carries higher copyright risk and may be harder to defend as copyrighted.
AI-assisted content: the more human creative input, editing, and unique material added, the stronger the claim to authorship.
Proprietary data: avoid uploading confidential or sensitive customer data to third-party models; deidentify if analysis is essential.
Fact-checking: always verify model-provided facts and sources; models can hallucinate plausible-sounding but incorrect information.
Transparency: consider noting AI assistance in content if desired (e.g., “created with AI assistance”) and consult legal counsel for specific situations.
Decisions
Use GenAI as a drafting/ideation/editor tool, not as a final publisherial source.
Prefer free models to start; upgrade to paid only when features (plugins, longer context) are necessary.
Do not upload confidential or proprietary data to third-party models unless deidentified and privacy is fully assessed.
Save and reuse good prompt templates; iterate and refine prompts over time.
Open Questions
How to quantify the minimum human editing needed to secure copyright protection for AI-assisted works? (Legal gray area; consult IP counsel.)
For regulated industries, which specific GenAI controls and vendor contracts meet compliance needs? (Refer to trade association guidelines and legal advisers.)
How will evolving model training and regulation change acceptable practices for using public-model outputs in marketing?