AI models typically learn by memorizing patterns, then researchers bolt on reasoning as an afterthought. A new method called Reinforcement Pre-Training flips this approach—teaching models to think during basic training instead.
Good Morning from San Francisco, Meta's AI app turned users into accidental exhibitionists. 😬 People shared medical questions, legal
Meta users think they're chatting privately with AI. Instead, they're broadcasting medical questions, legal troubles, and relationship problems to the world through a public feed that many don't realize exists.
Apple announced new AI features at WWDC 2025 while most promises from last year remain unfulfilled. The company integrated ChatGPT into development tools, admitting gaps in its own capabilities. Will this pattern of overpromising continue?
💡 TL;DR - The 30 Seconds Version
🚨 Apple announced AI features at WWDC 2025 while most AI promises from iOS 18 remain broken or delayed.
📊 Apple Intelligence writing tools score 15-20% lower than ChatGPT on accuracy tests, while Siri improvements were delayed 8 months.
🔧 Apple integrated ChatGPT into Xcode because its own Swift Assist coding tool never launched despite being demoed last year.
💻 New AI features only work on Apple Silicon devices, excluding millions of older iPhones, iPads, and Intel Macs still in use.
🏭 On-device processing protects privacy but delivers worse results than cloud-based competitors like Google and Microsoft.
⚠️ Apple's pattern suggests waiting 6-12 months after launch for actually working AI features based on previous delivery delays.
Apple spent WWDC 2025 announcing new AI features while most of last year's promises remain broken. Siri still can't see what's on your screen reliably. Apple Intelligence writing tools produce mediocre results. The promised revolution feels more like a software update with bugs.
The company revealed a Foundation Models Framework that gives developers access to its AI models. It integrated ChatGPT into Xcode. It expanded Visual Intelligence beyond the camera. All impressive on paper. All likely to face the same delays and limitations that plagued previous announcements.
Apple's track record with AI launches suggests caution. The company promises big changes, delivers partial features, then quietly fixes problems months later while announcing the next wave of capabilities.
Apple Intelligence launched with iOS 18 in limited form. The writing tools work inconsistently. Siri improvements remain basic. The promised ability to understand screen content was delayed indefinitely.
Smart photo searches miss obvious objects. Email summaries often capture the wrong details. Notification summaries group unrelated messages. These aren't edge cases—they're daily frustrations for users who expected the AI revolution Apple marketed.
The company promised Siri would evolve into a capable assistant that could take actions based on what it sees. Instead, users got slightly better voice recognition and the same limited command set. Asking Siri to help with tasks in third-party apps usually results in "I can't help with that."
Apple's AI works best in controlled demos with prepared content. Real-world usage reveals the limitations quickly.
Apple's decision to run AI on devices sounds great for privacy. It creates serious performance issues. Complex AI tasks drain batteries faster. Processing large language models heats up phones. Older devices can't run the latest features at all.
The Foundation Models Framework only works on Apple Silicon chips. That excludes millions of iPhones, iPads, and Macs still in use. Apple's environmental claims about device longevity conflict with AI features that demand newer hardware.
On-device processing also limits capability. Cloud-based AI services can access vast computing resources. Apple's models must fit within the constraints of phone processors and memory. The trade-off shows in reduced accuracy and slower responses.
Apple integrated ChatGPT into Xcode because its own coding assistance remains inadequate. The company demonstrated Swift Assist last year but never released it widely. Partnering with OpenAI admits Apple can't match established AI capabilities.
The same pattern appears across Apple's AI strategy. Visual Intelligence now uses ChatGPT for complex questions because Apple's models can't handle them. Image Playground gained ChatGPT styles because Apple's image generation produces limited results.
These partnerships solve immediate problems while creating new dependencies. Apple touts privacy and on-device processing, then connects users to external AI services that process data in the cloud. The contradiction undermines Apple's core AI messaging.
Live Translation sounds impressive until you examine the details. Apple didn't specify which languages it supports at launch. Previous translation features started with a handful of major languages and expanded slowly.
The company claims real-time phone call translation, but achieving natural conversation flow requires perfect accuracy and minimal latency. Current translation technology struggles with accents, slang, and context. Phone call audio quality adds another layer of complexity.
On-device translation models are typically less accurate than cloud-based alternatives. Apple chose privacy over performance, which means users get worse results. The feature might work in controlled conditions but struggle with real conversations.
Visual Intelligence now analyzes screenshots instead of just camera views. This extends existing capability rather than creating new intelligence. The feature still depends on external services for shopping searches and detailed questions.
Apple's demo showed Visual Intelligence recognizing a jacket in a social media post. The same result could be achieved by cropping the image and using Google Lens. The added convenience doesn't justify calling it a major advancement.
The screenshot analysis runs on-device, then connects to external services for actual results. Apple gets credit for privacy-focused design while relying on other companies' AI capabilities for useful output.
The Foundation Models Framework gives developers access to Apple's AI models, but those models are demonstrably weaker than alternatives. Developers might prefer the privacy benefits, but they'll sacrifice accuracy and capabilities.
Apple requires three lines of code to access its models. That simplicity suggests limited functionality compared to more complex AI frameworks. Developers who need sophisticated AI capabilities will likely continue using cloud-based services.
The framework only works on newer Apple devices. Developers must choose between reaching all users with traditional approaches or serving only recent hardware with AI features. The fragmented user base complicates development decisions.
Apple's AI announcements follow a predictable pattern. The company demonstrates impressive capabilities on stage. Initial releases include basic versions of promised features. Full functionality arrives months or years later, if at all.
iOS 18 was supposed to bring AI writing assistance that competes with ChatGPT. The delivered features barely improve on autocorrect. Siri was supposed to become conversational and context-aware. It remains a glorified voice command interface.
WWDC 2025 repeated the same promises with new terminology. Visual Intelligence will understand everything on your screen. Translation will break down language barriers. AI will transform how you interact with devices. The marketing claims echo previous years' unfulfilled commitments.
Google's AI features work reliably across Android devices. Microsoft integrated ChatGPT into Windows successfully. Meta's AI powers popular social media features. Apple's competitors ship functional AI while Apple perfects demos.
The gap becomes obvious when comparing actual capabilities. Google Lens recognizes objects accurately. Google Translate handles real conversations. ChatGPT writes useful content. Apple's equivalents feel like beta software that shipped too early.
Apple's focus on privacy and on-device processing creates competitive disadvantages. The company chose principles over performance, which means users get inferior AI experiences compared to alternatives.
Apple emphasizes privacy to distract from capability gaps. On-device processing protects user data but delivers worse results than cloud-based AI. The company markets limitations as features.
Private Cloud Compute was supposed to solve this tension by providing secure cloud processing. The system remains largely theoretical. Apple hasn't demonstrated how it works in practice or explained why users should trust it over established alternatives.
The privacy focus also creates user confusion. Apple promotes on-device AI, then connects to ChatGPT for complex tasks. Users can't tell which features protect their data and which send information to external services.
Why this matters:
Q: Which devices can actually run Apple's new AI features?
A: Only devices with Apple Silicon chips. This includes iPhone 15 Pro and later, iPads with M1 or newer chips, and Macs with M1 or newer processors. iPhones older than the 15 Pro and Intel-based Macs are excluded, affecting millions of devices still in use.
Q: How accurate is Apple's current AI compared to ChatGPT or Google's AI?
A: Independent tests show Apple Intelligence writing tools score 15-20% lower than ChatGPT on accuracy benchmarks. Siri's contextual understanding remains significantly behind Google Assistant. Apple's photo search misses obvious objects that Google Photos identifies correctly about 85% of the time.
Q: When will the AI features announced at WWDC 2025 actually work?
A: Developer betas start immediately, public betas arrive next month, with full releases this fall. However, Apple delayed major Siri improvements by 8 months after iOS 18's launch. Based on past patterns, expect working versions 6-12 months after the initial release.
Q: Why does Apple keep partnering with OpenAI if their own AI is so advanced?
A: Apple's on-device models are limited by phone processor constraints. ChatGPT handles complex tasks that Apple's models can't manage locally. The partnership admits Apple can't match cloud-based AI capabilities while maintaining their privacy stance.
Q: Does Apple's privacy approach actually hurt AI performance?
A: Yes. On-device processing limits model size and accuracy. Cloud-based AI can access vastly more computing power and training data. Apple chose privacy over performance, resulting in AI features that work worse than competitors but protect user data better.
Meta users think they're chatting privately with AI. Instead, they're broadcasting medical questions, legal troubles, and relationship problems to the world through a public feed that many don't realize exists.
Disney and Universal sued AI company Midjourney for using their characters without permission to train image generators. It's the first major Hollywood lawsuit against generative AI, testing whether copyright law protects creators in the age of artificial intelligence.
OpenAI cut its o3 model prices 80% while launching o3-pro—a reasoning AI that takes over 3 minutes to respond but outperforms rivals on complex tasks. The move intensifies AI pricing wars and splits the market between fast chat models and slow thinking ones.
Publishers built their business on Google sending them traffic. Now Google's AI answers questions directly, cutting out the middleman. Major news sites lost half their visitors in three years. Some adapt with new revenue models, others fight with lawsuits.
Get tomorrow's intel today. Join the clever kids' club. Free forever.
