A Starter for AI Guidelines
đ» AT THE TOP:
AI tools are all around us. Some are really AI driven, and some have just added the letters to capture attention. News about what the technology is in full hype cycle right now. Letâs get clarity on how we should and shouldnât use these tools. These are early, reasonable guidelines that will continue to change as new information comes to light.
đ NUMBER 1:
When using public AI tools like ChatGPT, Bard, Bing, only input content that is publicly available.
THE BEST OF THE REST:
If you are in doubt about how information you provide to an AI tool might be used, assume it will be used to train a public AI model. Refer to Number 1 above!
If you need to use the tool for specific company business that will require you to share private company content, you must have a standing legal agreement/subscription with the service and its terms and conditions must be read and reviewed by you/your team and the Compliance team.
Never assume that an answer from a public AI model like ChatGPT, Bard, Bing is true. Fact check, fact check, fact check.
DO use these tools to frame out initial drafts of text, ideas, code but only if it doesnât require you to enter private information. Start with generic prompts and outputs then bring that output into company systems and refine from there.
Avoid passing off AI generated content as human generated to internal and external audiences. Consider providing a disclaimer if AI was used in content creation.
If you would like to experiment with an embedded AI tool (plugin to Word, Excel, etc.) please work with your Technology team to vet the tool before use.
đž What about AI image tools?
Use Generative AI Image tools like DALL-E and MidJourney for experimentation and internal use only.
Because the law is not clear yet, only use company paid stock photography services and/or royalty free stock images for public content.
đ Basic Terms 101:
Publicly Available - any content that is published online outside of a paywall. Public content that you can find using a search engine.
Generative AI - Artificial Intelligence technology that can create new content (text, code, images, audio, synthetic data) derived from data that has been used to train it.
Hallucination - When an AI chat returns content that is false/non-factual
LLM - a type of Machine Learning model that can perform various natural language processing (NLP) tasks. They are "large" because they contain billions or trillions of parameters.
Training Data - large bodies of content given to a language model so it can improve its ability to perform NLP tasks.
đŠGo Deeper:
https://www.techopedia.com/definition/34948/large-language-model-llm
https://www.mckinsey.com/featured-insights/mckinsey-explainers/what-is-generative-ai
https://witlingo.com/a-chatgpt-glossary/
https://jumpstory.com/blog/using-dalle2-images-for-commercial-purposes/
https://openai.com/policies/terms-of-use