Posts

Note: This is a slightly modified copy of my original WWDC25 note Key Takeaways 🔐 Passive quantum attacks are already relevant ⚡ Sensitive data needs quantum-secure encryption immediately 🛡️ iOS 26 provides automatic defence mechanisms 🆕 New CryptoKit APIs: Post-quantum HPKE (X-Wing), ML-KEM, ML-DSA Understanding Quantum Attacks While today’s encryption is secure against classical computers, quantum computing changes the rules. Even though sufficiently powerful quantum computers are not yet publicly available, attackers have already found ways to exploit the future power of quantum computing....

Note: This is a copy of the original WWDC25 note published by me Key Takeaways 🤖 Best for simple tasks; no facts, math, and code generation ✍️ Use clear prompts with commands, examples and length control 🛡️ Multiple safety layers: built-in guardrails, instructions, and input controls 🧪 Evaluation and testing required for better quality and safety Design for on-device LLM The on-device LLM has some limitations: Complex tasks Break down tasks into simpler steps to get good results Math calculations Use traditional (non-AI) code for calculations instead Code generation Avoid code generation prompts since the model is not optimized for code Factual or world knowledge Don’t rely on the model for facts and recent events Can still be used in games and scenarios where the accuracy of the output is not too important Hallucinations Use guided generation (see doc:WWDC25-286-Meet-the-Foundation-Models-framework) to improve response reliability Hallucination example (model thinks plain bagels have toppings): Prompting best practices Use length qualifiers, such as paragraph or word count...

The Backstory Recently, I encountered a challenge while working on an iOS project that required a specific functionality that proved to be highly complex to implement. My search led me to a Java library that had the exact features I needed. However, there was one major roadblock: it was only available in Java. At first, I tried to find a native alternative that could be easily integrated into my iOS project via SPM, CocoaPods, or other dependency managers....