If anyone still has any of the old wiki content, I'd done this for a number of topics, and it might help kickstart things.
ES seeks Knowledgebase editors!
- Thread starter neptronix
- Start date
Once there is KB content, it would be helpful if the XF auto-helper-finder thingyat the bottom of a page would detect what someone is typing up and start showing them suggestions of KBs that have the info they're after, before they actually publish the post. Anything it shows that would take them somewhere *must* open in a new tab so they don't lose their post or place where they were.
We could actually do this search part, but the search may need to be in it's own box and rely on you only entering key terms because the search engine in zeropress cannot filter out all the regular words yet.
I would let you preview the article in a modal window before posting it, so we're not navigated away from the page and we don't have a browser tab to close. In fact, you could click outside the modal window and it will close, so, pretty convenient.
v-- what a modal window looks like:
Not sure if I was really clear, but I meant like when a newbie is posting the 1993375th version of the "how do I...??" thread,
and you know how XF shows you a box below the thread title you're typing if you start typing a thread title in the line at the top of a forum list? 
Or when you're reading a thread, then at the bottom it shows "similar threads"
well this would be for while someone is typing up the OP of a thread, before htey post it, suggesting KB articles that might be related. It would need to be more than the XF version does, because that often shows totally irrelevant stuff just because a word matches, but I don't know how that all works at the backend. Might be too hard or resource-intensive.
Would be nice if it also happened when KB editors are making an article, to help them find and link related content; this version could suggest other articles in tlittle boxes it shows inline with what the KBE is typing up. They can leave them there, and they'll stay as part of the artilc,e or drage them around to be in other places in the text. Or they can close them with an X and discmiss the suggestion.
I know that in website design it's possible to force any click on a link / button / etc to always open a new tab, even if the user doesn't do it, as i have run across sites now and then that say on the button or link that "this will open the content ina new tab or window", so it should be psosible to do that in XF or the KB, preventing the need for any modal dialogs or losing data from the tab being recycled for the new content instead of just leaving the post or article entry window opne as-is in it's own tab. But I have no idea how to do that...just that someone does.
A-DamW
10 kW
Definitely, a random collection of links would not work.
AW, that's unfortunately a tall order, technology wise, to come up with some search engine that handles natural language instead of just thinking about keywords.
We can't reuse the technology behind Xenforo here, it's awful in a lot of dimensions.
Let's say, making it ultra convenient and good and fast to slap a KB article into a reply is something that's a must have, i think we revisit the idea when/if we have content.
We can't reuse the technology behind Xenforo here, it's awful in a lot of dimensions.
Let's say, making it ultra convenient and good and fast to slap a KB article into a reply is something that's a must have, i think we revisit the idea when/if we have content.
Last edited:
It's just an idea...I'm full of them, practical or not.
marty
10 MW
I got so many ideas. Can't remember them all.It's just an idea...I'm full of them, practical or not.
Should write them down before I forget.
Last edited:
I was poking around in the KB and clicked on Add and was surprised that it created a new Article (in the Newbie section). I wasn't sure why, since I'm not an editor. I couldn't figure out how to delete it.
Anyway, are there any priorities when it comes to populating the KB? I was thinking if I were an editor, I probably wouldn't know where to start. If there's a way of creating a wish list, there are a few subjects that come up repeatedly that take a lot of typing. Some of them are subjects I have no interest in, like lighting, that can either be the main subject of a thread, or embedded often in conversion threads as one component. While I have interest in resolving the more ebike related technical issues, I don't have the patience to address lighting, but if there were a comprehensive KB article that could be referred to, that would be super helpful (to me, being selfish here) to provide to the poster.
Anyway, that's just one example where I think a definitive guide would be a worthwhile investment. Any thoughts on creating some priorities for content? I might be the only one that avoids certain subjects, due to the higher level of effort required and where interest is low.
Also, there are some members here who have successfully upgraded their factory ebike, and have gone through the light conversion as well. That input, or review of the article, would be valuable as well.
Anyway, are there any priorities when it comes to populating the KB? I was thinking if I were an editor, I probably wouldn't know where to start. If there's a way of creating a wish list, there are a few subjects that come up repeatedly that take a lot of typing. Some of them are subjects I have no interest in, like lighting, that can either be the main subject of a thread, or embedded often in conversion threads as one component. While I have interest in resolving the more ebike related technical issues, I don't have the patience to address lighting, but if there were a comprehensive KB article that could be referred to, that would be super helpful (to me, being selfish here) to provide to the poster.
Anyway, that's just one example where I think a definitive guide would be a worthwhile investment. Any thoughts on creating some priorities for content? I might be the only one that avoids certain subjects, due to the higher level of effort required and where interest is low.
Also, there are some members here who have successfully upgraded their factory ebike, and have gone through the light conversion as well. That input, or review of the article, would be valuable as well.
Last edited:
I made you an editor against your will for kicks hoping you'd poke around 
Deleting an article isn't super intuitive right now but this is how you delete one:
I would say i'd like to see newbie articles but if you have something else that lights your fire, go for it.
Deleting an article isn't super intuitive right now but this is how you delete one:
I would say i'd like to see newbie articles but if you have something else that lights your fire, go for it.
Low-and-Slow
10 mW
I've been dragging my feet on my offer to help. I joined the forum as a complete newbie to assisted riding and it has been a great source of information. My primary interest is recumbent trikes, assist kits that meet specific purposes and all the little bits and pieces that are missing in "complete" kits that are available. After waiting to see the KB up and rolling it looks like a good time to see if I have anything to offer.Knock knock!
As far as we know, the software works, is adequate for the task, is easy to use, and i've done a good enough job introducing and promoting it.
But 7 months go by and we only have a few drops of information in there
So far it's quite a flop ( as far as being useful to members )... meanwhile i have another 2 organizations using this tool and loving it.
Is there something that:
- i am missing
- i am doing to make contributing difficult
- i could be doing to encourage people to contribute
Open to your brutally honest feedback. Give it to me.
Would love whatever contributions we can get.
We have these major improvements planned:
- article editor should go from annoyingly buggy when doing advanced formatting, to quite smooth
- comment notifications for authors and comment number display in article list
- make summaries way better, right now they do a very bad job of tempting the user to read an article
But these improvement have been stalled due to lack of interest in contributing. ( we also have much other stuff to do )
I'd consider yours, or anyone else's future contributions a vote of confidence that we should continue improving the software from 'good enough for ES' to 'awesome for ES'
Give our knowledgebase a spin, i could use feedback!
We have these major improvements planned:
- article editor should go from annoyingly buggy when doing advanced formatting, to quite smooth
- comment notifications for authors and comment number display in article list
- make summaries way better, right now they do a very bad job of tempting the user to read an article
But these improvement have been stalled due to lack of interest in contributing. ( we also have much other stuff to do )
I'd consider yours, or anyone else's future contributions a vote of confidence that we should continue improving the software from 'good enough for ES' to 'awesome for ES'
Give our knowledgebase a spin, i could use feedback!
Last edited:
I copied and pasted some stuff into the battery section. Any revue, vetting, edits, corrections submitted to the comments will be appreciated and can/will be incorporated.
Whew, gotta take a nap now.
My will to improve the software is trending up!
Mind if i edit directly for some clarity improvements? I want to see how the revisions work
If not, i can copy to another article and you can make the final editorial decision ^_^
Mind if i edit directly for some clarity improvements? I want to see how the revisions work
If not, i can copy to another article and you can make the final editorial decision ^_^
Go for it!
Hi, I just came across this thread and I am loving it; a real knowledge base is something that I have always wished was here. I would definitely be interested in contributing; I haven't had time to be extremely active on the forum, but I am good at writing and have a good amount of knowledge /experience with VESC, brushless motors (a common newbie question is the difference between battery amps and phase amps, battery voltage, motor voltage, duty cycle, etc., it took me a bit to wrap my head around all that years ago when I was starting), battery design (also something that took me a while to grasp when I was new), and all the other random pieces of knowledge that I have accumulated but can't list off the top of my head.
Writing samples: Unfortunately I haven't been extremely active in contributing back to the forum, but here's my posts/replies: https://endless-sphere.com/sphere/members/ljbred08.113472/#recent-content
I have two current builds myself which I really need to write up a build log on at least one of them (that one I took a lot of picture/video on hoping to find the time to put it together someday).
As for writing outside of Endless Sphere, I have a few academic papers from a class a while ago (I am younger than most here
): Essay, Essay
and a short story I wrote last fall: Essay
Not technical writing, but I'm sure I easily pass the bar of high school level and my spelling/grammar are definitely above average.
Other skills:
I am good at CAD (3D printing) and coding:
Writing samples: Unfortunately I haven't been extremely active in contributing back to the forum, but here's my posts/replies: https://endless-sphere.com/sphere/members/ljbred08.113472/#recent-content
I have two current builds myself which I really need to write up a build log on at least one of them (that one I took a lot of picture/video on hoping to find the time to put it together someday).
As for writing outside of Endless Sphere, I have a few academic papers from a class a while ago (I am younger than most here
): Essay, Essayand a short story I wrote last fall: Essay
Not technical writing, but I'm sure I easily pass the bar of high school level and my spelling/grammar are definitely above average.
Other skills:
I am good at CAD (3D printing) and coding:
Actually that's pretty trivial nowadays with the current tech that has evolved with the AI boom; I have done it myself. I have no idea how this forum is built or who the infrastructure maintainer is, but I could definitely help with anything on the coding side, and I agree that it will need some work to make sure that the knowledge base is usable if we get anything written in it.
Happy to hear it!
First, as far as writing skills and knowledge level go.. you have access to edit now:
I believe my controller terms 101 article covers the basics okay, it's open to additions to make it better.
Knowledgebase
On the controller front, we could use:
- advanced version of controller terms 101
- VESC for dummies ( as it applies to the kinds of vehicles we build here )
Or whatever else you're interested in contributing, even if it's just an outline, or filling out an existing one.
We'd love contributions to the software; we plan it to be an open source project once it exits incubation here at ES. We could make the code available on github if you'd like to contribute there.
Our software is written in PHP and we do our best to keep it simple so it's hackable.
But it's a little on the complex side anyway since it offers to many conveniences to the end user.
( making things easy is hard! )
It's built on the Zerolith framework, which in a nutshell, is like a series of helpers for writing raw PHP.
A developer can learn and use it in less than a week by reading the docs.
I am the guy who built 90% of it, and so i could be advised on how it works if anything is unclear.
AI works pretty well at writing raw PHP, but good taste, understanding existing code, and some years of PHP experience are required to work on it. Luckily, the knowledgebase is not critical infrastructure for ES yet, so it's forgiving to coding mistakes.
Let me know if that aspect is of interest. We tend to do the hard parts first and leave the easy stuff for later. So there is lots of bite sized improvements to make in addition to the big ones i mentioned earlier.
I see you had a bad time uploading images. This should 'just work', but doesn't once you have a large amount of images, it seems.
It is not worth it to fix the existing wysiwyg editor. We need the new wysiwyg editor, which should be a big holistic upgrade for editing in itself.
And also, the fact that you don't have a profile set produces a busted image as your user pic.
Back to the drawing board, but thanks for helping us kick the wheels
It is not worth it to fix the existing wysiwyg editor. We need the new wysiwyg editor, which should be a big holistic upgrade for editing in itself.
And also, the fact that you don't have a profile set produces a busted image as your user pic.
Back to the drawing board, but thanks for helping us kick the wheels
Last edited:
Checking in.
I have a Nvidia 5090 and checked out the new open weights LLM ( AI ) model - GLM 4.7 Flash.
It produces surprisingly good knowledgebase article summaries, and it's pretty fast at it.
Very good mix of smarts to efficiency.
It can take a few generations to land a good one, but each generation takes under 10 seconds.
I was impressed with the test outputs.
But yeah it can write way better automatic summaries than the current ones, which are just the top part of the article with the html stripped out which has turned out to be wack on ES
What do you editors think about this workflow:
The article summary entry is no longer just a text box, but a text box + a magic button that reads the article and proposes a summary, and you can accept it and it will fill the summary box for you, you can roll the dice for another summary, or cancel and go manual.
Affordability wise, the AI power cost is tiny due to the small size of knowledgebase, and it's also cheaper to build than a better version of our automatic summary generation. So, the finance department is ok with it
I have a Nvidia 5090 and checked out the new open weights LLM ( AI ) model - GLM 4.7 Flash.
It produces surprisingly good knowledgebase article summaries, and it's pretty fast at it.
Very good mix of smarts to efficiency.
It can take a few generations to land a good one, but each generation takes under 10 seconds.
I was impressed with the test outputs.
But yeah it can write way better automatic summaries than the current ones, which are just the top part of the article with the html stripped out which has turned out to be wack on ES
What do you editors think about this workflow:
The article summary entry is no longer just a text box, but a text box + a magic button that reads the article and proposes a summary, and you can accept it and it will fill the summary box for you, you can roll the dice for another summary, or cancel and go manual.
Affordability wise, the AI power cost is tiny due to the small size of knowledgebase, and it's also cheaper to build than a better version of our automatic summary generation. So, the finance department is ok with it
Last edited:
New article created about controller programming:
Knowledgebase
Interestingly i did not have issues putting multiple images in.
Both copy and paste and drag and drop worked fine.
@ljbred08 i'm wondering if you can try again and if you fail, let me know if you see any error messages. I also want to see these attachments, it would be nice if you could post them here.
Knowledgebase
Interestingly i did not have issues putting multiple images in.
Both copy and paste and drag and drop worked fine.
@ljbred08 i'm wondering if you can try again and if you fail, let me know if you see any error messages. I also want to see these attachments, it would be nice if you could post them here.
Checking in.
We are working on this implementation for the new WYISWYG editor as we speak.
I recently found a way to overstuff a bigger AI model into our existing hardware, and the result is something smart enough to find obscure bugs in our code with good accuracy ( 3 of 4 are actual problems that a human hasn't caught yet )
It is also faster and more CPU efficient in addition to being smarter.
Let's check what it looks like if we ask it to summarize our intro post into a maximum of 300 characters:
Knowledgebase
I used a simple prompt: 'please summarize this article'.
This does look like AI generated text, but it does a good job of summarizing. I think if we add some writing guidelines, it could be both more on point and more on tone. It also only took 0.2 seconds to generate.
Also, this is a programming specific model. There are models that specialize in creative or technical writing which may be better.
I think it is possible to go from 'good enough' to 'good' at a minimum.
Right now i don't use LLMs to generate text often. I hate reading LLM produced text. The only thing i've used it for is documenting software, and each time, i end up editing the result by hand. The human : machine hybrid process produces good results though. Nobody can tell i'm cheating on my homework so far.
If/when we add this feature then i would want our editors to hand edit the results until we have AI good enough to nail it a majority of the time. But even still, it should require a human lookover and approval.
We are working on this implementation for the new WYISWYG editor as we speak.
I recently found a way to overstuff a bigger AI model into our existing hardware, and the result is something smart enough to find obscure bugs in our code with good accuracy ( 3 of 4 are actual problems that a human hasn't caught yet )
It is also faster and more CPU efficient in addition to being smarter.
Let's check what it looks like if we ask it to summarize our intro post into a maximum of 300 characters:
Knowledgebase
I used a simple prompt: 'please summarize this article'.
This does look like AI generated text, but it does a good job of summarizing. I think if we add some writing guidelines, it could be both more on point and more on tone. It also only took 0.2 seconds to generate.
Also, this is a programming specific model. There are models that specialize in creative or technical writing which may be better.
I think it is possible to go from 'good enough' to 'good' at a minimum.
Right now i don't use LLMs to generate text often. I hate reading LLM produced text. The only thing i've used it for is documenting software, and each time, i end up editing the result by hand. The human : machine hybrid process produces good results though. Nobody can tell i'm cheating on my homework so far.
If/when we add this feature then i would want our editors to hand edit the results until we have AI good enough to nail it a majority of the time. But even still, it should require a human lookover and approval.
Last edited:
Side note, i see E-HP is a bit active in the KB
@E-HP , i noticed that you wrote some more advanced versions of my 101 articles. We have some overlapping content, and you seem more interested in writing about the nitty gritty details whereas i am trying to dial it down to the beginner level as much as possible.
Maybe we could pair up and you could write the 102 curriculum and i'll keep at filling out the curriculum for 101. What do you think?
@E-HP , i noticed that you wrote some more advanced versions of my 101 articles. We have some overlapping content, and you seem more interested in writing about the nitty gritty details whereas i am trying to dial it down to the beginner level as much as possible.
Maybe we could pair up and you could write the 102 curriculum and i'll keep at filling out the curriculum for 101. What do you think?
Sure. I almost forgot I added that stuff. The structure wasn't obvious to me, so after I sorted alphabetically and then by date, I just figured I’d start typing and sort it out later.Side note, i see E-HP is a bit active in the KB
@E-HP , i noticed that you wrote some more advanced versions of my 101 articles. We have some overlapping content, and you seem more interested in writing about the nitty gritty details whereas i am trying to dial it down to the beginner level as much as possible.
Maybe we could pair up and you could write the 102 curriculum and i'll keep at filling out the curriculum for 101. What do you think?
Just FYI. We are making good progress on getting the WYSIWYG editor working.
This editor uses a different storage format. So for articles created before we switch to this editor, will open in the old editor when you hit 'edit'.
Like the icons a lot better on this new one.
This editor uses a different storage format. So for articles created before we switch to this editor, will open in the old editor when you hit 'edit'.
Like the icons a lot better on this new one.
I'm able to load a 120b model instead of a 80b AI model now.
Here's new summaries generated from this article:
Knowledgebase
In this case, the model had to be directed what to focus on with a prompt. Otherwise it got fixed on the fact that most of the text relates to the software itself ( this makes sense since 66% of the text focused on the software that runs it )
The language on these is a better. It sounds less like ad copy & follows the original text much closer and more literally
We should have 4x the video RAM soon and so it would be possible insert additional instructions for 'tone' ahead of the request.
I think it is possible to get right, we'll develop this auto-summary feature in the near future
Here's new summaries generated from this article:
Knowledgebase
In this case, the model had to be directed what to focus on with a prompt. Otherwise it got fixed on the fact that most of the text relates to the software itself ( this makes sense since 66% of the text focused on the software that runs it )
The language on these is a better. It sounds less like ad copy & follows the original text much closer and more literally
We should have 4x the video RAM soon and so it would be possible insert additional instructions for 'tone' ahead of the request.
I think it is possible to get right, we'll develop this auto-summary feature in the near future
Last edited:
Checking in.
We have TipTap integration ~60% done.
I have onboarded another part time developer to start an internal cleanup round of our KB software and then get to adding some features relevant to ES.
AI hardware is here and i now have 128gb of vram.
The extra hardware was increasing in price faster than we can get funds to help defray the cost, so i just maxed out my credit cards.
We can run a 197B model ( Stepfun 3.5 Flash ), and so far, it kicks ass and reminds me of a much bigger AI model!
The result of this hardware upgrade was WAAAAAAAAAY better than expected
It successfully generated 201 lines of documentation for some of my software, being asked to copy the tone of another handwritten documentation for another piece of software. The result was better than the by-hand version, very factual, and would require minor editing to pass as if it was written by me.
So i gave it a harder challenge, to summarize this article:
Knowledgebase
For generating summaries, the model's default tone is flat and factual. Not a surprise for a model that specializes in coding.
It needed a more specific instructions.
It followed my instructions very well. The tone is very close to ideal.
We just jumped from 'good enough with some editing' to 'good with minor editing'.
The answer to:
- can AI help us write more code and make bigger impact for ES?
- can AI help us with a few boring jobs?
Appears to be solid yes
We have TipTap integration ~60% done.
I have onboarded another part time developer to start an internal cleanup round of our KB software and then get to adding some features relevant to ES.
AI hardware is here and i now have 128gb of vram.
The extra hardware was increasing in price faster than we can get funds to help defray the cost, so i just maxed out my credit cards.
We can run a 197B model ( Stepfun 3.5 Flash ), and so far, it kicks ass and reminds me of a much bigger AI model!
The result of this hardware upgrade was WAAAAAAAAAY better than expected
It successfully generated 201 lines of documentation for some of my software, being asked to copy the tone of another handwritten documentation for another piece of software. The result was better than the by-hand version, very factual, and would require minor editing to pass as if it was written by me.
So i gave it a harder challenge, to summarize this article:
Knowledgebase
For generating summaries, the model's default tone is flat and factual. Not a surprise for a model that specializes in coding.
It needed a more specific instructions.
It followed my instructions very well. The tone is very close to ideal.
We just jumped from 'good enough with some editing' to 'good with minor editing'.
The answer to:
- can AI help us write more code and make bigger impact for ES?
- can AI help us with a few boring jobs?
Appears to be solid yes