Some things generative AI tools are really useful for:

• Translation
• Transcription
• Suggesting code in many programming and scripting languages
• Summarizing or pulling specific information from long, already-existing documents or collections of documents
• Assistance with classifying or archiving information for easy retrieval
• First drafts of low-priority correspondence
• Thinking through complex ideas or narratives
• Making entertaining images to help visualize things (but not to share publicly as embarrassing “boomer images“)

What generative AI tools really are not useful for:

• Answering questions using information drawn from the open web: the answers are often vague or hallucinated, at times hilariously so, and always require further verification

You’d think that Google and other companies rushing to “put AI in everything” would know this. Anyone who has used even the best LLMs for more than a few hours has had the model tell them outrageously wrong things. It’s puzzling that Google put “AI overviews” at the top of their trillion-dollar company’s results pages when the tech is plainly not ready for that purpose yet.