How Can I Contribute OR Why YOU Will Drop Ubuntu Touch Entirely
-
@projectmoon said:
@oldbutndy some are valid. Some are technically correct, but don't matter. Others are matters of opinion. But to me it mostly reads like complaining. Iterating a lot of problems (real or supposed), and using a sycophantic AI to write long winded arguments.
Instead of AI generated complaints, maybe take action?Maybe I misunderstood when I read some @grenudi comments. I thought AI was used to try to summarize a large number of hours of research into something more concise.
My own use of AI, in asking google to estimate how long to write a kernel driver for a sub section of hardware (after reverse engineering it), was to get a sense of just how much effort that is. (No idea here, since I am hardware guy with some embedded controller coding, and data sheets were always available). So, when I multiplied all the various manufacturer sub sections by all the hours required (AI est.) to reverse engineer and code kernel drivers, the effort required seemed to be hundreds or thousands of times as large as the current method using halium layer.
Summary: only way it (native component drivers in the kernel for a whole device) will ever happen is if OEM's WRITE those drivers.
Translation (for past devices) = NEVER.
Future: Maybe (brax 3 ? or others ?). < update here, I don't think Brax 3, since I think it requires halium.
So, from what I understand of all this so far, is UT developers are making this work the ONLY way practically possible. Best way architecturally ? does not matter. No other way.
Future: when multiple millions are using UT ... -
Long term, you are correct.
BUT (and sadly), I just asked (to get an idea of effort) AI 'how long to reverse engineer cellular modem chip to create linux kernel driver' ?
Answer was several thousand hours over 6 to 18 months.
If anyone anywhere has a better answer than that, AI or not, please tell me.
What that means to me is, given the circumstances, and the continuous change in devices, and ALL the sub modules in our modern cell phones, we will probably NEVER have native Linux kernel for OLD phones.
Best chance near term might be something like Brax Phone 3 (a NEW phone) - IF they really have native Linux Kernel drivers for all components.I havent thought of that, thats a good point and a food for future thoughts. Glad to see constractive conversation around here. Thank you
-
I thought AI was used to try to summarize a large number of hours of research into something more concise.
Yes, and to counter prove the points made, fact check facts, and so on, working on it as a project in claude and not just "make a huge paper on how ubuntu touch is bad. And make it look cool and stuff. Proceed", as I'm getting impression that some here think that this is the only way to use llms, and that they all are the same, that because free gpt now produces halucinations that means pro version claude is the same and all of them are. By the way have you noticed that they have "downshifted" free use of geminy in recent days, just like it was with gpt ? Now its much harder to get something useful out of it, it even messes up given queries confusing words contexts
-
As I've said, if ai produces slop for you, you are using it wrong, or using a wrong ai
-
-
where expert volunteers could start with upstreaming, and so forth.
Thank you for the suggestion. I'll think about it. I'm currently crafting such comprehensive roadmap. Dont know if anything useful will come out of it and it will require 5 times, or even more claude tokens, than the current project took, which already was a lot and currently I can not afford more.
-
where expert volunteers could start with upstreaming, and so forth.
Thank you for the suggestion. I'll think about it. I'm currently crafting such comprehensive roadmap. Dont know if anything useful will come out of it and it will require 5 times, or even more claude tokens, than the current project took, which already was a lot and currently I can not afford more.
I would say doing something constructive is better than complaining, at least.
-
I asked claude for ESP32 project help (about 10 days ago, for first time ever) & it went reasonably well.
So, yesterday I asked if it could help code a UT app to talk to that ESP32 project over wifi.
Claude said better solution was webserver on ESP32, then access through any browser from any phone.It gave a summary of good sounding pro / con's, with that same 'confidence' all AI's seem to have whether they are 100% right, 100% wrong, or anywhere in between !
Do you use claude for coding, as well as analysis / summary of things ?
-
I would say doing something constructive is better than complaining
@projectmoon exactly! Glad for this mutual understanding
-
Do you use claude for coding, as well as analysis / summary of things ?
Yes. I want to advise you to own that execute button on any ai, especially claude, he loves to jump straight into execution, create ton of how to, code snippets fixes and so on, burning your tokens like a teenager his life. Own execution button, I mean that I spicifically instruct it to outlay the whole plan of actions and how he is going to execute them, relay the whole project plan, my intent in .md artifact, then I read the artifact and instruct it on more specifics, he relays it back extended, and so on, untill we have as detailed plan as possible of what i want from him and how i want him to do it, and what are expected outputs, what and how to validate and so on. This is especialy helpfull to bootstrap a new proqect and template architecture, and you do not want him to create bigillion of project files and constantly adjust the mess afterwards, easier to do it in one go (it will ask a lot of continue tool usage), and more conservative on your token balance.
and another tip - dont trust their context always instruct it to write researched, gathered, googled info in facts artifacts (I use .md at the moment) in the project and consult with them. Because I've noticed they tweak things on those llms all the time and they can be very lazy, stating from time to time something with 100% confidence when in reality it never fact checks, googles those claims(for that I have permanent instruction in memory to always factcheck by trusted resources its claims and place a disclaimer, always, if it cannot be done).
just wanned to share my exp. Maybe you already know all that so sorry for the unasked advise. Maybe it worth a post somewhere but im pretty sure there is plenty already about it, and its much easier to ask the thing itself
how to speak to it in the most productive way and dont burn tokens for nothing -
@oldbutndy said:
Do you use claude for coding, as well as analysis / summary of things ?Yes. I want to advise you to own that execute button on any ai, especially claude, he loves to jump straight into execution, create ton of how to, code snippets fixes and so on, burning your tokens like a teenager his life. Own execution button ...
... Maybe you already know all that so sorry for the unasked advise.THANK YOU. I REALLY appreciate your insight on this. Always open to advice. I did not know it.
I asked Claude, it answered. Got C++ code working (TFT LCD, WiFi NTP, RTC, SD micro, SDI temp sensor, LiPo monitor, etc). But I had no idea how many tokens I used.
Then yesterday: Slow/No reply - then something like: " you have to wait until 8 pm." - HARD STOP - Ohhh, OK.
Hello! It looks like you're interested in this conversation, but you don't have an account yet.
Getting fed up of having to scroll through the same posts each visit? When you register for an account, you'll always come back to exactly where you were before, and choose to be notified of new replies (either via email, or push notification). You'll also be able to save bookmarks and upvote posts to show your appreciation to other community members.
With your input, this post could be even better 💗
Register Login