AI For The Blind Lesson 1 How does AI work?

Here is the transcript, if you want to contact Shawn email [email protected] or call/ text 817-659-9353 "Good evening, I’m Shawn, and tonight we’re going to talk about how AI works with language and image models." "When you type a question into something like ChatGPT, the computer is really good at looking for patterns in language. For example, AI understands simple phrases like ‘Mary had a little lamb’ because it’s seen that pattern so many times. It knows what comes next by analyzing how words commonly follow each other." "The same applies when you generate images with AI. The model looks at millions of images it’s been trained on, identifying basic elements like, ‘Oh, that’s a dog’ or ‘That looks like an apple.’ It uses those patterns to build the image you’ve asked for." "Now, this is where it gets tricky—sometimes AI hallucinates. I once took a picture of a garage with horizontal wood slats, and the AI thought it saw stairs leading to a door. It’s fascinating because it really believed it was right until I started asking more questions, like, ‘Could that be siding?’" "When generating an image, AI relies on basic elements. For example, it knows what a landscape looks like because it’s been trained on millions of landscapes. But when it comes to image descriptions, it might sometimes miss obvious details or focus on the wrong parts until you ask the right questions. That’s why when I generate images, I often double-check descriptions with tools like Aira Access, asking it to 'describe everything in full detail' and 'look for any unusual AI renderings.'" "In the end, AI is incredibly powerful, but it’s still learning. Whether it’s text or images, you have to guide it by asking the right questions and refining your prompts." Support AI For The Blind. by contributing to their tip jar: https://tips.pinecast.com/jar/wespeak This podcast is powered by Pinecast.