Since it seems AI is pretty good at reverse-engineering stuff like this, is there any educational material on how to use it for that purpose? Seems like it could really help port things like postmarketOS to new devices (and improve support on existing ones)?
I have claude code hooked up to deepseek, I hooked up my spare cheapo android tablet, installed adb and fastbook with my package manager and asked the AI to jailbreak the tablet.
It discovered the tablet was running a unisoc t606, found a CVE from a couple years ago, and unlocked the bootloader for me. I was the meat puppet holding the "volume up" button and plugging in the usb cable a bunch of times. Like most of my experiences with this stuff, it was pretty eerie.
Next step for me is to attempt mainline linux, there seems to be some postmarketOS devs playing with it. We've probed most of the tablet's hardware except the exact display.
So this is using deepseek-v4-pro / flash and using claude code as the harness right? That's really good that it works. I'm pretty impressed with how v4-pro is doing, sadly there's no subscription packages so I'm not using it too much. (Wafer used to offer it but now they don't).
I have some experience on this and could make an article if you are interested.
The key is to have downstream sources and be very very conservative with the AI, slowly build step by step.
You also have to know C and have a spider sense of what's acceptable or not.
Another key is to ask for approval before editing any source with a patch of what it intends to do. This way you can judge what it wants to do and ask for a double check of the patch. Go quality over quantity.
This isn't web frontend with Tailwind, you have to be very strict and somewhat knowledgeable. Nobody can use AI to write kernel code without some good low level and engineering knowledge.
I completely agree, this is not the place to let AI blindly edit kernel code. The useful approach is to use it conservatively: understand the error, compare against downstream sources, propose a small patch, review it, test it, and then move one step further.
I’d be happy to work together on an article or guidance document, where to start, how to approach debugging, what to never let AI touch blindly, and how to build confidence step by step. That could help others avoid a lot of mistakes and maybe give a second chance to other devices.
Please do write an article! I've wanted to get into reusing old android hardware for quite some time now, but never knew where to look for good instructions to get started. Especially PostmarketOS seems very interesting, but rather underdocumented in some places.
I will then, didn't know it would be interesting for other people.
As for PostmarketOS, I've built my own tooling scripts around it to make it easier to build patches, debug hex variables, switch between downstream/mainline and rebuilding everything with a single command. (Unrelased yet though).
I find their tooling okay for a release for end-users but a bit clunky for debugging.
There are things I will just not bother to learn. I can either not do them, or let AI do them for me. There are things I can do for myself, but can't be bothered. I can either not do them or let AI do them for me.
I prefer spending my time doings I actually want to do. Let the machine do the boring things.
You can be dedicated to Biomedical Medical science and your whole world may revolve around it. You may be the smartest person in any given room, although sometimes it might not be worth learning something else given your time constraints or energy constraints.
If said Biochemist needed to write a simple Python script, why would he bother learning Python, setting up the .env and debugging when an AI could do it and he could go back to doing whatever he was doing?
I bet you don't understand any of the maths/physics that I work with. You see, I prefer not to waste my time with things that machines can do faster. Some people like the illusion of mastery though.
It helps for fuzzing, maintaining and is actually a great help for seniors, maybe not for the ones who don't care for the project and publish slop.
It could now actually help a lot in some ways not just coding though but things surrounding project management.
No it doesn't. It helps lazy people. It helps Lazy seniors, it helps lazy project managers. You have this ass backwards. It helps everyone who is Lazy.
