Raspberry Pi CANBUS Adventure

Disassembly, Reassembly, Tools and devleopment. Going deep with Hardware and Software.
User avatar
Meistro
Posts: 35
Joined: Fri Nov 03, 2023 7:42 am
cars: many

Re: Raspberry Pi CANBUS Adventure

Post by Meistro »

start here and ask questions. https://chat.openai.com/ I've been messing with this one for a while. I even pay for it to get more out of it. It started as me asking it to write Saturday Night Live skits based on my input, lol. Then I wondered if it could write macros for me to use in my cad software. I was hooked after that. Sometimes it needs to be set straight but I have learned so much about python and javascript from this thing. I had it calculating wind loads for a project I was working on. I tried a few times and just asked, "How can I hack into my cars ECU to change the vin number" It didn't want to have anything to do with that :( You gotta lead into the topic gently. If that makes sense.
User avatar
Gatecrasher
Posts: 273
Joined: Sat Apr 25, 2020 6:09 am

Re: Raspberry Pi CANBUS Adventure

Post by Gatecrasher »

Your AI buddy is doing the same thing it's image generating cousins do. It's the equivalent of 6 fingers on a hand. The individual parts make sense, but the whole is wrong.

DID read is 02 1A 90. I have no idea where it's getting F1 from since all mode 1A requests are 2 bytes. The byte count is wrong anyway, so it's not even correct in its own incorrect context, if that makes sense.

I respect what you're doing, but you need to take a much more methodical approach to this.
User avatar
antus
Site Admin
Posts: 8253
Joined: Sat Feb 28, 2009 8:34 pm
cars: TX Gemini 2L Twincam
TX Gemini SR20 18psi
Datsun 1200 Ute
Subaru Blitzen '06 EZ30 4th gen, 3.0R Spec B
Contact:

Re: Raspberry Pi CANBUS Adventure

Post by antus »

Yep Ai is a tool, not going to say it isn't, but its essentially got a huge database of internet posts, and we are not told which ones or from where, also certain books. Then it applies a form of statistical analysis over that to create a model which can give you the probable next word to your prompt. Its not actually thinking, in any way, and it doesn't know when its wrong. It'll generate text that looks like it make sense, but it's only true for 80% of it. You can't tell the difference until you dig deep in to the output yourself. It's good to write a template, or to get you started, or give you a direction to research further but it's a mistake to assume its right about everything it says. You can use it, and you'll spend time debugging and fixing just like you would writing your own code. It's another angle with another tool. I tried to get it to write some CRC32 code for 68k processor. It used opcodes that didn't exist on that platform (specifically xor when it should have used eor), had some bugs in the routine, and missed a critical part of the algorithm (the polynomial). That's pretty typical. I have also read about a lawyer who used case references in a court of law to illustrate certain points. Except they didn't do their homework and the actual cases didn't exist even if the output sounded right. That one did not go down well. Another one was I was looking for software that could generate a trigger wheel I could take to a laser cutter. It named certain software, which looked about right. I installed it, no such functionality. I asked it which version of the software. It told me one, I tried that. Still no functionality. In then it said yes, I am wrong (I think that's the default response when you tell it that its wrong). They call it an AI hallucinating. https://en.wikipedia.org/wiki/Hallucina ... elligence)
Have you read the FAQ? For lots of information and links to significant threads see here: http://pcmhacking.net/forums/viewtopic.php?f=7&t=1396
Post Reply