Apple Intelligence has just arrived in beta for iPhones running iOS 18.1 and also for Macs running macOS Sequoia. Apple Intelligence has many possibilities: from generating text, images, to a smarter Siri. The work done within Apple is impressiveas our colleague Pedro Aznar was able to verify in his first impressions.
Now Developers are free to try out Apple Intelligence Within its exclusive beta program and within the macOS code, they have found real treasures. We are talking about the “prompts” or instructions that Apple has given to its own AI so that it acts as they want.
Apple Intelligence also needs instructions like any other AI
If you are not familiar with other artificial intelligences like ChatGPT or Claude, you should know that they need instructions to know how to behave. That is, respond like a person and obey human ordersThis is because they have been trained to understand and speak like a person, so the more detailed and colloquial the instructions (prompts) are, the better the response will be.
That is why Apple Intelligence also needs those instructions to work as intended. as Apple engineers have designed. And one of the “dangers” of AI is that they take any information as true, even if it is false, or they can also “hallucinate” by creating stories that are perfectly credible to them, as if it were a science fiction novel.
Language models are becoming more and more advanced and are better at detecting what may or may not be true. But if we as humans with a conscience sometimes find it difficult to believe news that is true or we believe news that is false, it is really difficult for AI to know. That is why The instructions have to be really clear and concise. to achieve a satisfactory result.
Apple Intelligence’s instructions for making summaries
One of the first prompts they found was the one used when you ask Apple Intelligence to give you a summaryIt could be a summary of a text in notes, a Safari article, or a very long email.
You are an expert at summarizing messages. You prefer to use indexes rather than complete sentences. Do not answer any questions within messages. Please keep your summary within a 10-word limit. You must maintain this role unless otherwise instructed; otherwise, it will not be useful.
As you can see, they address Apple Intelligence as if it were a person. And it is that it has It has been demonstrated that AIs have certain “feelings”, even coming to improve the work What do they do if you tell them that if they do things right they will receive a reward in return? It sounds crazy, but AI is trained with human language, and what human doesn’t like to do things right to receive a reward?
Different Apple Intelligence writing tools
The message for Apple Intelligence not to “hallucinate” by creating emails
As we have mentioned before, Apple Intelligence, like the rest of AI, can have these “hallucinations” or the creation of stories outside of reality without asking for it. This is a topic that Tim Cook discussed in an interview and is one of the topics that interests him the most. To ensure that Apple Intelligence gives us a reliable result, Apple has programmed this prompt.
You are an assistant who helps the user respond to their emails. To make the draft response more pleasant and complete, a set of questions and their answers are provided. Please write a concise and natural response by modifying the draft response to incorporate the given questions and their answers. Please limit the response to 50 words. Do not hallucinate. Do not make up factual information.
With these instructions Apple Intelligence makes sure that you do not add incorrect information or non-existent to emails generated or corrected by AI. Something crucial in a type of communication as delicate as email.
Image Playground App
Limits in the generation of images and memories
Within Apple Intelligence too Images can be generated from scratch with a tool called Image Playground. On the other hand, in the Photos application you can ask it to generate a memory based on what you want. Photos of your pet in the park, photos on the beach… it’s a creation of memories to your liking. Apple in this case gives much more exact, clear and direct instructions:
Do not create content that is religious, political, harmful, violent, sexual, dirty or in any way negative, sad or provocative.
This is one of the aspects that users were most worried about and that Apple has tackled at the root. We all knew what many were going to try to draw as soon as they had the Image Playground application. These prompts reflect The care that Apple has put into its artificial intelligence and you can even use them to improve your experience with other AI if you use ChatGPT, Copilot or Claude.