You guys know the joke about how pets and their owners start to resemble each other?
The core functionality is simple:
Automatically, upon each payment, add the expense to my app
Update an Apple Watch complication with the % of my monthly budget spent
Categorize the purchase for later analysisCan someone enlighten me? I don’t understand why you need AI for this in the first place.
Reading the payment and turning it into structured data
Dozens of apps and services do that already and and have for years before AI existed.
Yea, but those are all using heaps of proprietary heuristics.
The beauty of LLMs and one of their most useful tasks is taking unstructured natural language content and converting it into structured machine readable content.
The core transformer architecture was original designed for translation, and this is basically just a subset of translation.
This is basically an optimal use case for LLMs.
Quite obviously not the optimal use case. “The tensor outputs on the 16 show numerical values an order of magnitude wrong.”
That’s the hardware issue he was talking about, it has no relation to the effectiveness of the usage of the LLM. It sounded to be mostly a project he was doing for fun rather then out of necessity
Grok says it’s right so it must be 🤤
tl;dr: because of LLM’s.
Certainly his use of LLM was stupidly egregious, but he found that even by those standards the math results underpinning the LLM were way off.
He combines LLMs with numbers and wonders why this does not work? Under which rock does he live?
I think you missed the point of his post. His issue is that the numeric operations the phone executes to run the LLM is producing garbage. Arguably this could break all kinds of neural networks, such as voice transcription. He’s not complaining that the LLMs are themselves unable to properly perform math.
He also had it work on a Mac, an iPhone 15 and an iPhone 17. Only his iPhone 16 got the internal LLM state wrong. It’d be interesting to know how a failure like that happens. Presumably most iPhone 16s have a working NPU. Apple would surely want to get to the bottom of this but I doubt they would be open about their findings. Maybe they do know but the solution is ‘buy new iPhone’.
I wish you better luck the next time you try to read.
Under which rock does he live?
Under the rock where reading comprehension exists apparently.
Where he was prompting for “What is 2+2?” to the LLMs, the accuracy of the answer was immaterial. At that step he was comparing two systems and simply needed a static question to give both system to compare the internal processes to determine why they arrived at different outputs (or a what appeared to be race condition/infinite loop for one) when the result should be identical to both irrespective of how right or wrong the answer is to the prompt. The LLM answer from the LLM could have been “ham sandwich” and it still would have served his purposes.
I went with quantized Gemma
Well, was it quantized in a way that iphone 16 supports?
Often it’s the quantization where things break down, and the hardware needs to support the quantization, can’t run FP16 on int8 hardware… And sometimes the act of quantization can cause problems too.
And yeah, LLMs are likely going to be very hit or miss anyway.
deleted by creator
I wonder if this hardware issue is specific to HIS iphone 16 pro max, or ALL iphone 16 (pro, max)s
Given he apparently found a bunch of forum posts of people complaining about erratic behaviour so it may be more widespread
Might be helpful to have a reproducible test case for it.
