GenAI Diary (extension) -- Enhanced productivity using Apple Intelligence

 Last update: Tuesday 2/11/25

Use case: iPhones for enhanced productivity via generative AI ➡ Do not use Apple Intelligence until it can process text with far greater competence than in iOS 18.3 The rationale for this prohibition is explained in this note. iOS 

Return ➡ Diary

Computer savvy iPhone owners who want to use generative AI to increase their productivity at work and/or at home should stop using Apple Intelligence immediately. They should use high quality chatbots instead, like ChatGPT, Gemini, Claude, or DeepSeek Chat.

Note: iPhone owners can still use the visual enhancements offered by Apple Intelligence, e.g., custom emojis, editing photos, creating images.

Note (2/11/25): The text processing skills of iOS 18.3 represent a substantial improvement over iOS 18.2, but still lag behind the skills of ChatGPT. Whereas iOS 18.2 had language skills that roughly matched an average 10th grader, iOS 18.3's skills roughly match an average college sophomore, better, but not good enough for computer-savvy college graduates, the intended audience for this blog

Unfortunately, the current version of Apple Intelligence has serious deficiencies in processing text and transcribing a user's speech into text during dictation, specifically:

• Apple Intelligence is much less capable of correcting a user’s errors in spelling, grammar, and capitalization than OpenAI's ChatGPT and Anthropic's Claude. 
  
  
• During dictation, Apple Intelligence often misperceives the words that were spoken, specifically the component syllables (phonemes) of the words the user actually said.
  
  Of course this deficiency will be experienced by some users more than others, e.g., users from different racial groups, age groups, regions, etc. This  variance is probably due to Apple's coming so late to the AI game. It has not had as much time to train its models on the voices of as many different groups as were involved in the training of ChatGPT and Claude. 
  
  
• In dictation, its pre-trainied models (the "P" in ChatGPT) did not include some of the most commonly used AI names and AI technical terms in artificial intelligence, a topic likely to be mentioned in notes written by users of GenAI tools. For example, Apple Intelligence knows how to spell "ChatGPT" with appropriate capital letters, but stumbles over terms like "chatbot", and "OpenAI"
  
  
• Apple Intelligence is far less effective in recognizing common idioms 
  
  
• Apple Intelligence is a "slow learner". Users can correct its mis-corrections again and again and again, but it will keep on mis-correcting text that needs no correction.
  
  
• Apple Intelligence often chooses the wrong homonym during dictation because it does not give sufficient attention to the surrounding text.
  
  
• And finally, a vexing deficiency is making  "corrections" that introduce grammatical errors into the user's text

BackToTop