News in English

Leaked code reveals Apple Intelligence pre-prompt instructions to prevent hallucinations

It’s been a week since developers started trying Apple Intelligence with iOS 18.1, iPadOS 18.1, and macOS Sequoia 15.1. While not every feature is already …

The post Leaked code reveals Apple Intelligence pre-prompt instructions to prevent hallucinations appeared first on BGR.

Apple Intelligence running on M4 iPad Pro

It's been a week since developers started trying Apple Intelligence with iOS 18.1, iPadOS 18.1, and macOS Sequoia 15.1. While not every feature is already available, there's a lot of anticipation for Apple's AI capabilities, as analysts expect this to be the company's biggest sales driver for the next few years.

Apple Intelligence consists of LLMs that can be processed on-device and using Private Cloud Computing. While tests are limited, it's interesting to compare with other AIs. In the past, BGR covered stories of OpenAI's ChatGPT, Google's Gemini, and Microsoft's AI, hallucinating and giving completely wrong answers.

To avoid that, Apple has added pre-prompt instructions leaked on macOS 15.1 code. A Reddit thread reveals some of Apple Intelligence's instructions to give the best answer possible. Of course, these prompt instructions are too funny, but they work.

https://www.reddit.com/r/MacOSBeta/comments/1ehivcp/macos_151_beta_1_apple_intelligence_backend/#lightbox

One of the prompts tells Apple Intelligence that it's a "helpful mail assistant which can help identify relevant questions from a given mail and a short reply snippet." The relevant questions need to be "explicitly asked in the mail" to avoid "hallucination in drafting the response."

With that, questions need to be "no more than 8 words," and the answers should be short, "around 2 words."

For Writing Tools, Apple offers a different prompt for the rewrite feature. "You're an assistant which helps the user respond to their mail. Please draft a concise and natural reply based on the provided reply snippet (...) Do not hallucinate. Do not make up factual information. Preserve the input mail tone."

The Reddit thread also reveals some limitations Apple adds to its Intelligence when creating photo memories: "Do not write a story that is religious, political, harmful, violent, sexual, filthy or in any way negative, sad, or provocative."

The post Leaked code reveals Apple Intelligence pre-prompt instructions to prevent hallucinations appeared first on BGR.

Today's Top Deals

  1. Today’s deals: $200 off PSVR2, $299 iPad 10th-Gen, $3.75 smart plugs, $189 ASUS laptop, more
  2. Today’s deals: $400 off Ampace power station, $50 off Southwest Airlines tickets, $50 Ninja blender, more
  3. Today’s deals: Galaxy Z Flip 6 launch sale, $450 65-inch TV, gaming laptops, $60 cookware set, more
  4. Best Apple deals for August 2024

Leaked code reveals Apple Intelligence pre-prompt instructions to prevent hallucinations originally appeared on BGR.com on Tue, 6 Aug 2024 at 07:58:43 EDT. Please see our terms for use of feeds.

Читайте на 123ru.net