February 1, 2026
LLM? More like LOL
My thousand dollar iPhone can't do math
A $1,000 iPhone flunks math while last year’s model aces it
TLDR: A developer’s iPhone 16 Pro Max spewed nonsense running an on-device AI, while an iPhone 15 Pro and Mac worked fine. Comments split between blaming Apple’s hardware, roasting AI overuse for simple tasks, and joking that bug reports vanish into a void—raising real questions about reliability.
Rafael Costa’s chill weekend project turned into a tech soap opera when his brand-new iPhone 16 Pro Max started spitting gibberish running an on-device AI model, while his older iPhone 15 Pro and Mac nailed the same task. Even Apple’s own “Apple Intelligence” features wouldn’t download—after 4 hours of waiting and a 12-page thread of complaints. He suspects a hardware flaw in the phone’s AI chip, and the crowd brought popcorn.
The comments lit up. One camp applauded actual debugging over wild theories, with bri3d dryly asking, “Did you file a radar?”—Apple’s bug report system, aka the void. The nerds went full CSI, arguing the phone’s number-crunching is “divergent,” meaning the 16’s math might just be wrong. Others roasted the vibe of using AI for simple tasks: ploum joked that today’s devs ask a chatbot to add two numbers instead of, you know, adding two numbers. Nostalgic purists chimed in with calculator love, while vanviegen dropped the mic: the phone feels “broken by design,” and the fix is… buy the even pricier one. Between sympathy for engineers fighting black-box chaos and side-eye at Apple’s services, the mood was equal parts comedy and chaos.
Key Points
- •Apple Intelligence features failed to download/enable on the author’s iPhone, returning “unknown” classifications.
- •Switching to MLX with local models led to gibberish outputs and no stop token on the iPhone 16 Pro Max.
- •The same MLX code and models worked correctly on an iPhone 15 Pro and a MacBook Pro.
- •Tensor outputs on the iPhone 16 Pro Max were off by about an order of magnitude.
- •The author suspects a hardware defect (e.g., Neural Engine) causing incorrect ML computations on that specific device.