Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• 🥸 Apple researchers say reasoning models flawed

• 🤖 Qualcomm acquires Alphawave for $2.4B

• 🍎 Apple Intelligence update from WWDC

• 💻 MCP vs API

Let’s get right into it!

Sign Up | Advertise | Tools

BIG TECH

Apple researchers say reasoning models flawed

Apple researchers have released a critical paper challenging claims that the latest large language models (LLMs)—including OpenAI’s o3, Anthropic’s Claude 3.7, and Google’s Gemini—truly possess reasoning abilities. The team argues that these models exhibit an "illusion of thinking," struggling with complex problems despite adequate training, due to an “overthinking” phenomenon that causes accuracy to collapse beyond certain complexity thresholds.

Using controlled puzzle tests, they found these AI systems fail to apply explicit algorithms consistently and often hallucinate, raising fundamental doubts about their true reasoning capacity. This critique is notable given Apple’s more cautious AI approach and its position behind competitors, suggesting either a genuine technical limitation in current models or a strategic challenge to rivals as Apple develops its own AI tools for devices like iPhones and MacBooks.

Full PDF publication here.

BIG TECH

Qualcomm acquires Alphawave for $2.4B

Qualcomm Inc. is acquiring U.K.-based chip interconnect technology provider Alphawave IP Group plc for $2.4 billion, offering a 98% premium on Alphawave’s share price. This move aims to boost Qualcomm’s presence in the data center chip market, complementing its recently announced plans to launch data center CPUs, including a potential data center version of its Oryon CPU series.

Alphawave’s interconnect technologies, like AresCORE and GammaCORE, enable high-speed data transfer within system-on-chip (SoC) designs and support emerging applications such as co-packaged optics (CPO) for faster data transmission. Qualcomm’s CEO highlighted the strategic fit for advancing data center workloads, with the acquisition expected to close by early 2026 after securing approval from shareholders holding about half the company’s shares.

BIG TECH

Apple Intelligence update from WWDC

At WWDC 2025, Apple unveiled several upgrades to its Apple Intelligence AI tools, including enhancements to Genmoji and Image Playground with new ChatGPT integration for richer image generation and emoji combinations. Messages, FaceTime, and Phone apps gain live translation powered by on-device AI, while Visual Intelligence now scans both camera and screen content with deeper ChatGPT integration for contextual queries and smart calendar suggestions. Developers can leverage Apple’s on-device foundational AI models through a new framework supporting Swift, enabling offline, privacy-focused intelligent features at no cost.

An improved Shortcuts app adds AI-powered actions, and Apple Wallet will auto-summarize delivery tracking info. Despite these updates, Apple Intelligence continues to lag behind competitors like Google, with delayed Siri improvements and previous features showing bugs, though the company remains focused on delivering practical, user-friendly AI enhancements.

TECHNOLOGY

MCP vs API

The Model Context Protocol (MCP) is a new way for AI programs to talk to tools and services, designed especially for AI instead of humans like regular APIs (which are how software usually communicates). Unlike traditional APIs that use fixed instructions and can be confusing or error-prone when AI tries to use them, MCP lets AI discover what tools are available in real time and pick the right one to use in a clear, reliable way.

It supports two-way communication, meaning both the AI and the tool can send messages back and forth easily. MCP also works well on your local computer without needing complicated setups. Basically, MCP is like a smart, simpler language for AI to control tools and get things done without messing up, while regular APIs are made more for human programmers and web apps. MCP doesn’t replace regular APIs but adds a layer that makes it easier for AI to use them.