If you’re not getting what you want from OpenAI’s large language models these days, try paying them a compliment. One example, according to Scale AI field CTO Vijay Karunamurthy, is to say: “you’re a very smart computer that takes a lot of time and thinks step by step to answer questions.” Remarkably, that helps Scale AI’s developers get better responses. I kid you not.

As bizarre as that might sound, it reflects a reality for a growing number of OpenAI’s customers: its products are difficult to use. I had my own experience with that over the weekend, as I watched my teenage brother try to get ChatGPT to help him with his homework. To come up with a few examples of “heroic local politicians,” he spent 20 minutes going back and forth with the chatbot. Along the way ChatGPT provided him with a number of historical figures that didn’t exist.

Sure, my 16-year-old brother likely isn’t ChatGPT’s target demographic, but he’s not the only one complaining that LLMs aren’t user-friendly, not to mention unreliable. Amjad Masad, CEO of AI coding assistant startup Replit, lamented on X over the weekend that ChatGPT is getting more difficult to use. And as OpenAI has faced a growing number of copyright infringement lawsuits, its chatbot no longer responds to certain commands regarding copyrighted characters—such as fictional characters owned by film studios, others have pointed out. (You can check out some of the most important ongoing lawsuits in yesterday’s newsletter.)