There's your on-device AI, bro
you need a shitload of vram to do useful models, only basic shit can work on-device, normies are connected to the internet 24/7 so this isn't relevant anyway
>>106466919Apple's built-in AI apps are dogshit.Luckily there's apps that you can run image/video generators (SD1.5,SDXL,Flux,WAN) and LLM local models directly on it.The iPhone 15 Pro and 16 series are about the same speed as a 3050 with 8GB VRAM, pretty impressive for something that uses 4W max.https://apps.apple.com/us/app/draw-things-offline-ai-art/id6444050820
What local AIs can I sideload on Android? Anything available on F-droid?
>>106468933Qualcomm did some SD1.5 demos back in 2023, but they have yet to release anything.MLC Chat seems to be the main one for LLMs.Problem is that the models have to be compiled specifically for Snapdragon, while on iOS you just run the safetensors file from civitai or gguf straight from huggingface
>>106468882what's a "shitload"?
>>106469163like 12GB+, and then you need additional ram because the rest of the operating system and apps still have to function. The 1-4GB models can work on fancy android phones with like 12-16GB ram
>>106469315Solution status?
>>106468922Pretty sure Apple decided that current generative AI is shit and can’t be used in a “finished” product, so they’ve only used it for stuff like built-in transcription, object identification, OCR in images (this one is particularly useful) and then slapped on the ChatGPT integration to shut retarded investors up.
>>106469386Why are you pretending apple never shipped an unfinished product?
>>106469402They have, I just think they try not to if they can avoid it. They market themselves on “it just works”, so they’re generally careful to maintain that image.
>>106468882I asked my watch to convert temperatures and Siri couldn’t because I was in the sauna and out of phone rangeSo I had to open PCalc and type numbers in like an animal
