> I created a RAG agent to hold most of GCP documentation (separate download, parsing, chunking, etc)
If you share the scripts to gather the GCP documentation this, that'd be great. Because I have had an idea to do something like this, and the part I don't want to deal with is getting the data
> This is a OTA vehicle update. It has the ability to update the infotainment, ECU, ECM, TCM, and BCM. Multiple manufacturers have been able to release recalls that fix major vehicle defects (safety, reliability, and performance). *That wouldn't be possible without OTA updates that update core vehicle computer systems.*
Why wouldn't it be possible without OTA? It would just require someone to go somewhere, or do something, to get this installed.
While their assumption is incorrect, your conclusion is incorrect.
From their own essay, the plugins are simple. They even give examples. They claim maintenance, sure. But they also claim the core API is stable and shouldn’t see changes.
Which is it?
----
I talked about them getting paid above, that’s fine. The rug pull and then coming up with these arguments are the parts that don’t sit right with me.
> For v1, we moved a handful of convenience plugins into Datastar Pro.
and
> Nothing you can build was taken from you; we set a support boundary
If it was available on core, it was supported by them. If they moved it to Pro, isn't still supported by them?
Not sure what the 'support boundary' is. If they didn't want to provide official support for it, wasn't 'core' the better solution for them anyway? Wouldn't pay require them to officially support it?
----
The ability to build is separate from the convenience of prebuilt. It is paywalling things. This is like saying, 'you can send electrical pulses to your computer, no need for an OS or tooling'.
If everything is achievable through the same api, then the plugins wouldn't do anything. If they simplify things, then do they do add something, convenience. This is what plugins do, which they say aren't needed? But if they're not needed, what's Pro for?
Yes, it's a 501c3... it's still commercial since they're selling...
If it's stable, no v2, plugins aren't needed, it's a 501c3, there's no shares, equity... what's the point of Pro? "The goal is to fund the work and draw a clear support boundary," What are they funding?
By adding a Pro subscription, what's the incentive to work on core?
---
As an outsider it just looks as a way to justify Pro. But it's not a technical explanation or explains the maintenance policy.
I completely agree. The "nothing was lost because you can still build it yourself" argument could just as well be used to make the entire thing proprietary. You still have your Turing Machine, so you didn't lose anything!
I don't have a big problem with the "paid bonus features" business model in general, but when you're removing features from the open-source version, at least be honest about it. And it's open source: if it breaks you get to keep both pieces, so where's the support burden supposed to come from?
I have no problem with paid features. I have no problem with them making money. I have no problem with them monetizing it this way. I have no problem with the 501c3.
I think that moving things from free to paid is them forcing the hand of users to pay, which is sweeping the rug under them. This is more on _how they did it_, more than the fact that it's paid.
I just don't like how they did it and it's not a good argument to justify it.
> By adding a Pro subscription, what's the incentive to work on core?
The article literally says that they consider core complete after numerous rewrites and optimizations. The whole point of a plugin architecture is that extensions don't have to touch core source code, so development on Datastar, open or closed source, is all about the community of plugins.
"You don't need plugins to do useful work with Datastar", and "You don't need these plugins for most work" are both perfectly reasonable interpretations.
The data attributes they moved to Pro was mostly a result of watching folks use Datastar and seeing the anti-patterns develop. Those anti-patterns have been moved to Pro.
> If it's stable, no v2, plugins aren't needed, it's a 501c3, there's no shares, equity... what's the point of Pro? "The goal is to fund the work and draw a clear support boundary," What are they funding?
I assume it is because charitable organizations need accountants and other things (along with all the other stuff like web hosting and the like).
If they make less than $50,000 USD, they need to file the 990-N postcard, online.
It's less than 10 fields? Things like name, ein, fiscal start, end, etc.
If they make more than $50k, they can fill the EZ form. Sure, hire an accountant if you want. Most is how they earned the money, assets, expenses, where they spent the money. They need to declare officers too.
If they earn I think it's $200k, then they need to fill the IRS 990 form. Sure, get an accountant.
There's another requirement thrown in the above if they have more than X in assets...
Ironically Non-profits have to show profits. We don't particularly want devs money as much as Teams/Enterprise to pay for tooling. Inspector make life easier and more stuff (like Stellar is in route).
> As you increase the size of the input data, the accuracy gradually decreases.
Interesting.
On your section "Limitations and Areas for Further Study",
What I'd be curious on future work would be,
- changing the order of the data on each table type
- changing the order of the questions
I'm curious to know if what it fails is the same, if it changes depending on the location, if it's a bias.
Is it always a specific question? Is it always a specific value? Is it always question #x (or around question #x?). Does it tend towards x or y on types of questions?
LLMs have documented position biases, with skew towards first and last. This is strongest in messages due to system prompt + current question training data, but it's present in list data in general.
Exactly. But the papers I’ve seen, the tests are done based on answers being multiple choice usually.
Where do you eat?
A) floor
B) table
C) dirt
In this case, the questions asked have an answer. The bias would then be on the order of the input data. It’s different enough that it triggered my curiosity.
If you share the scripts to gather the GCP documentation this, that'd be great. Because I have had an idea to do something like this, and the part I don't want to deal with is getting the data