I use both ChatGPT and Claude. I strongly suggest taking a look at Claude, it surprised me with it's breadth, precision and less tendency to hallucinate.
I have been watching your videos for a little more than a year and tbh tuning for me is so fucking confusing. actually watching this video of yours helped me actually find answers to what my issue was and you fact checking the chatgpt helped me understand the pids and what they do even more. a lot of my quads are micros and whoops that do not have blackbox.
i used claude to find better kv and pids for 4" LR and it was more or less ok, it was suggesting even playing with TPA. well Vlaude made way more really good points
yes the speedbee f405 v3 has sd card issues, can't use it. It's one of the main reasons I really love your tune by feel vids. "The SpeedyBee F405 V3 flight controller shares the SPI2 bus with its Blackbox logging and analog OSD. This means that enabling Blackbox logging can sometimes cause the analog OSD to disappear. Disabling Blackbox logging restores the OSD."
How much do you charge for tuning a quad remotely? I have built a 3.5” Volador with SpeedyBee f405 stack, RCInpower GTS V3 1804 (2450KV) motors, DJI O3, ELRS and flying with GEPRC 1100 HV batteries. Getting lots of prop wash and when using a lot of throttle getting some wobble. Just want it flying right. It’s currently on Betaflight 4.5 and have the Supafly preset on it.
So I tried with chat GPT, but went with a different approach. I entered in the Specs of my quad, Like motor size/KV weight of the quad Props specs and told it what firmware I had. and it spit values right out. I used them and to my surprise they worked. maybe it was luck but all the oscillations i was getting are gone now and the quad fly way better. also tried with my whoop and it flys better as well. maybe there something to this.
speedybee f405v3 has a hardware conflict with the OSD and SD card. you cant have both, so thats why his SD card black box isnt working.. A while back, I fed my black box to CHATGPT, but it understandably did nothing. I didnt try hard
My v5 frame was trash. Every standoff needed to be tapped. Holes are so small the screws thread into the carbon and can damage threads on inserts during assembly attempts. Arm x plate rattles insode frame no matter how much you tighten it. Definitely will never waste money on another one.
All I can say is yes, if you know how to ask it the right question. I had a 3-in toothpick and I did not know what to do cuz the pid tune were messed up but I asked it very specific questions while I was tuning and it definitely helped me out. It may not be perfect but it gets you up and going and flying.
Searching for the betaflight documentation or your videos on PID Tuning will be faster and more useful for your brain than trying to get an AI toddler to give useful answers.
Tuning a quad with help of an AI model would probably work if you would train a model with lots and lots of evntually hand selected flight logs of quads that fly from very bad to exceptionally good. Then the model may be able to find some correlations between the curves and the tuning values and it may be able, if you then give it the flight log of your quad to tell you which values to set to achieve no noise and nice smooth curves and whatever else is needed that a flight log looks like the flight log of a perfectly flying quad. That's what such a model can work with. All answers like what we've seen here are BS because the model has not the slightest understanding what all that really means To be honest: It simply said "just play with all the pids and filters in small steps." Unfortunately it takes a real professional AI engineer to train a model like in the way I described it in a few simple words. I said just give it lots of data but it has to be trained with goals and even to find the concrete goals may be very complex. But maybe ! it's even easier. - Maybe it is sufficient to just give it the data of 100000 or more flight logs as so called "embeddings", then give it the flight log of the bad flying quad in combination with the question to adjust all tuning parameters in a way that they may correlate to smooth curves and as less noise as possible etc. With this approach getting a really good answer the real work would be narrowed down to just "LLM prompting" - First explaining the LLM with examples in it's system-prompt what these flight logs mean and how they can be read and "understood" and then developing the correct question to ask it, to get the best tuning-parameter values as an answer. Maybe that would already work? - But even if not: Maybe it would work if all data was classified with quad size, weight, etc. and those details of the quad in question would also be given in the question to the LLM. - If you would also add categories like "quad for cinematic filming", "freestyle quad", "longrange quad", etc. you would possibly even be able to ask it for a perfect configuration for your quad of one of that categories. Maybe ChatGPT has already been trained on thousands of betaflight logs? - Who knows? Did you try to pass it a log, explain how to read it and ask it then? But it's absolutely essential that you first explain to it what kind of data you are giving to it! Ah, I forgot: Because the flight log data is not what we see in our log viewer, the format of the dataset itself has to be explained to it!
I use AI Claude to tune my filter and I give a picture of the log data to track whether it is excessive or lacking. However, it needs to be improvised again with our knowledge, don't take it raw
and if you don't know that he's lying, and you can know if you know, and if you know then why ask. and if you don't know and you ask and he's lying then you can do something weird)))))
![Lp. Сердце Вселенной #60 РОЖДЕНИЕ ЛОЛОЛОШКИ [Финал] • Майнкрафт](http://i.ytimg.com/vi/YoR0pAV9FVQ/mqdefault.jpg)
I use both ChatGPT and Claude. I strongly suggest taking a look at Claude, it surprised me with it's breadth, precision and less tendency to hallucinate.
I have been watching your videos for a little more than a year and tbh tuning for me is so fucking confusing. actually watching this video of yours helped me actually find answers to what my issue was and you fact checking the chatgpt helped me understand the pids and what they do even more. a lot of my quads are micros and whoops that do not have blackbox.
Yeah refined and detailed questions the bigger the better make it more accurate
i used claude to find better kv and pids for 4" LR and it was more or less ok, it was suggesting even playing with TPA. well Vlaude made way more really good points
yes the speedbee f405 v3 has sd card issues, can't use it. It's one of the main reasons I really love your tune by feel vids. "The SpeedyBee F405 V3 flight controller shares the SPI2 bus with its Blackbox logging and analog OSD. This means that enabling Blackbox logging can sometimes cause the analog OSD to disappear. Disabling Blackbox logging restores the OSD."
ahahaha. I thought at first... we've come to this. that's cool.. and then I sit and laugh))) cool)
In Inav I have a problem with 10 inch Quadcopter tuning, for 10 inch Quadcopter predefined tune not available in Inav
see my website
How much do you charge for tuning a quad remotely? I have built a 3.5” Volador with SpeedyBee f405 stack, RCInpower GTS V3 1804 (2450KV) motors, DJI O3, ELRS and flying with GEPRC 1100 HV batteries. Getting lots of prop wash and when using a lot of throttle getting some wobble. Just want it flying right. It’s currently on Betaflight 4.5 and have the Supafly preset on it.
So I tried with chat GPT, but went with a different approach. I entered in the Specs of my quad, Like motor size/KV weight of the quad Props specs and told it what firmware I had. and it spit values right out. I used them and to my surprise they worked. maybe it was luck but all the oscillations i was getting are gone now and the quad fly way better. also tried with my whoop and it flys better as well. maybe there something to this.
The speedy bee flight controler
There are alot of f405 v3 with sd card issues...that was the big "change" to the fc on the v4 stack along woth bigger esc
Hey how about putting the blackbox files into ChatGPT. I tried but it said I had to get the paid version.
If it ever happens, I think its another few decades before AI can actually do coding, tuning, drawing etc as good as a regular professional.
try Grok :P
speedybee f405v3 has a hardware conflict with the OSD and SD card. you cant have both, so thats why his SD card black box isnt working.. A while back, I fed my black box to CHATGPT, but it understandably did nothing. I didnt try hard
ChatGPT is a language model, NOT a knowledge model. Ppl seem to don't get that. In terms of tech it just babbles along :)
I use my chatgpt to learn and tune my drone, this how i got more into the hobby and my personal trainer, definatelly the best trainer.
You need to have a talk with ChatGpt
agreed. Need to figure out how to talk to it.
@@uavtech 😅
Can chat GPT rewrite betaflight and make it work without bugs
submit and issue.
what bug? Must be minor?
Chatgpt is just very advanced search engine. Nothing else.
Its not even that. Techbros are trying to give it any application they can imagine to satisfy their sunken cost fallacy.
@MarkusRessel yeah could be. 😅😉
My v5 frame was trash. Every standoff needed to be tapped. Holes are so small the screws thread into the carbon and can damage threads on inserts during assembly attempts. Arm x plate rattles insode frame no matter how much you tighten it. Definitely will never waste money on another one.
4o is better, you know, you used up your 4o time. lol
you can just ask gpt to be brief and reply shortly, so it wont write an essay every time
All I can say is yes, if you know how to ask it the right question. I had a 3-in toothpick and I did not know what to do cuz the pid tune were messed up but I asked it very specific questions while I was tuning and it definitely helped me out. It may not be perfect but it gets you up and going and flying.
Until I see a definition of INTELLIGENCE I'm not like to use any AI.
Searching for the betaflight documentation or your videos on PID Tuning will be faster and more useful for your brain than trying to get an AI toddler to give useful answers.
Chatgpt told me to use kiss ultra on exp medium pids and it worked great, man thats easy 😁
That response is as garbage at it's Dshot Beacon answer. 🤣
The more people use it too the smarter it will get
ChatGPT is high 😂
Tuning a quad with help of an AI model would probably work if you would train a model with lots and lots of evntually hand selected flight logs of quads that fly from very bad to exceptionally good.
Then the model may be able to find some correlations between the curves and the tuning values and it may be able, if you then give it the flight log of your quad to tell you which values to set to achieve no noise and nice smooth curves and whatever else is needed that a flight log looks like the flight log of a perfectly flying quad.
That's what such a model can work with.
All answers like what we've seen here are BS because the model has not the slightest understanding what all that really means To be honest: It simply said "just play with all the pids and filters in small steps."
Unfortunately it takes a real professional AI engineer to train a model like in the way I described it in a few simple words. I said just give it lots of data but it has to be trained with goals and even to find the concrete goals may be very complex.
But maybe ! it's even easier. - Maybe it is sufficient to just give it the data of 100000 or more flight logs as so called "embeddings", then give it the flight log of the bad flying quad in combination with the question to adjust all tuning parameters in a way that they may correlate to smooth curves and as less noise as possible etc.
With this approach getting a really good answer the real work would be narrowed down to just "LLM prompting" - First explaining the LLM with examples in it's system-prompt what these flight logs mean and how they can be read and "understood" and then developing the correct question to ask it, to get the best tuning-parameter values as an answer.
Maybe that would already work? - But even if not: Maybe it would work if all data was classified with quad size, weight, etc. and those details of the quad in question would also be given in the question to the LLM. - If you would also add categories like "quad for cinematic filming", "freestyle quad", "longrange quad", etc. you would possibly even be able to ask it for a perfect configuration for your quad of one of that categories.
Maybe ChatGPT has already been trained on thousands of betaflight logs? - Who knows?
Did you try to pass it a log, explain how to read it and ask it then?
But it's absolutely essential that you first explain to it what kind of data you are giving to it!
Ah, I forgot: Because the flight log data is not what we see in our log viewer, the format of the dataset itself has to be explained to it!
lol
I use AI Claude to tune my filter and I give a picture of the log data to track whether it is excessive or lacking. However, it needs to be improvised again with our knowledge, don't take it raw
Try Bing Copilot
Folks skip to @7:12.
Skick to the regular methods of getting professional help from a human expert.
Wow.. I think the future is going to be ai flight controllers..in inav
and if you don't know that he's lying, and you can know if you know, and if you know then why ask. and if you don't know and you ask and he's lying then you can do something weird)))))