Understanding AI: Language Models and Fill-in-the-Blank Tests (Full Transcript)

Speaker 1: Quick language test, fill in the blank. Oh, wow, thank, question two. He's really tall, like six, last question. After six months, I finally said I love, let's see how you did. If your answers were thank you, six feet, and I love her, congratulations, you got a perfect score. But you could have also said thank goodness, six four, and I love him. Or maybe you said thank heavens, six two, and I love you. Regardless of what you said, if you got a perfect score, congratulations, you think like an AI. Or more precisely, an AI thinks like you. There's an AI out there developed by Google called BERT. BERT essentially learned English by taking thousands of tests with thousands of fill in the blank questions. After enough trial and error, it learned what sounded right when filling in those blanks. CHAT-GPT and GPT-3 work in a pretty similar way. Those models figure out what word is best to say next based on the words that came, you got it. Follow DeepGram for more AI content.