it now seems that ai agents picked up your documentation and are a lot more helpful in troubleshooting storj usage configurations.
i hope that this would get a lot of support burden off your already busy heads!
Oh, I dunno… people seem to be turning up on the forum with questions related to documentation that is out of date so I wouldn’t trust LLMs blindly.
A lot has changed in the last few months.
i also dunno, but at least codex saved me from bothering storj support at least 4 times.
Hi, may I ask where is storj AI? Never contact support before, inside storj dashboard or support page or somewhere?
it is not storj’s ai, it is 3rd party ai engines.
and they learn periodically by what they find on the web, including storj docs if they are published.
This is a wrong way to use ai. You don’t want it to try to remember what it read on the Internet. Instead you want to point it documentation and/or source, ask question, and have it read then and there and produce an answer. Whether it saw this in pre or post training is irrelevant.
well, my inquiries werent about the latest and greatest changes, just some noob user trivia that hadnt changed for quite a while.
and the starting point was that if i would have known what docs to look at, i would have probably solved half of my questions myself. 
my inquiries were mostly “integrate with this” and “integrate with that”.
the rest (a few) were “i want so and so, is it at all possible?”.
and btw, among all, moved the rclone mount that shares a bucket over lan into wsl onto ubuntu, with a samba share, now it works as i want and not impeded by windows network folder share limitations.
but the main comp remains on win because i need badly hyperv with gpup for some of my tech investigatory endeavours.
Right. Treating ai as all known deity does not work. It wil lie/be wrong confidently. It will recall what it saw more often in training. Popularity is a bad criteria for anything technical.
It is best used as a reasoning tool: look at this code/docs/specs, find and answer, and prove that it’s right.
One-shotting may work eventually (or if you spend considerable amount of time feeding context, which is no longer oneshotting) but we are not there yet. For this to work there shall be some way to share life experiences to have the same “defaults”.
When you give a vague task “like integrate this with that” there is a lot of ambiguity and it will make decisions prioritizing different things along the way than you would.
Another tangential thing is using ai for writing. It’s not only self defeating, but does not work in the first place. Does not stop some forum goers here from outsourcing their brain. /rant
And lastly, ai seems smart and capable only if you yourself are not expert. Keep that in mind. It elevates the baseline, but cannot be trusted where it matters. It’s not entirely ai fault. Humans a lot often wrong too in a high shot specialized areas.
well it saved me a lot of routine and i am pretty pleased overall.
most what i put codex at, is mad monkey coding by a ton a day and publishing beta releases sometimes even several times an hour.
there i decide what to do.
storj-related questions were only tangential to my main activity.
That’s for sure. Amount of tooling it created is astounding and does improve productivity. No more menial labour.
In my world (firmware and kernel drivers), that would have been quite frowned upon!
did a network-driver project some years ago so i know that.
now my mad development is at https://hotspot-observer.vv75.name/ where the focus is usefulness for the user and ui ergonomics first of all.
upd: for those that came there from here seeking fun, check out Compare Wi-Fi stability over time — narrated and sung tours and compare the 2 presentations.
dude from vietnam, did that make you just smile or laugh out loud?
i fell into deep thinking after doing that… not to mention that i was impressed myself.
had quite a dilemma which one to leave, and finally decided to leave both. bah. let everyone admire according to their preference.
took a considerable amount of “memorize …” rules to push into agents.md before it stopped misbehaving, now it works like a well-trained dog already.
and i added support for 4 major desktop linux distro families, and it is already not “several builds an hour” cuz it takes quite a laborsome time to test on all of them.
there are of course cases when ai can also be put onto testing, but it is not my case because the app specifics require a physical radio wifi adapter and not a virtual lan, and there is only one wifi usb adapter that i can plug into virtualbox vms, and testing is limited to only one at a time.
and on the storj side, storj still needs improvement because it is still not “work as if files were locally” even in native mode and not s3; lots of timeouts upon uploading files and reads may also get very delayed because of waiting for upload timeouts.
i sort of guess what you are referring to.
many people think that they could come out of blue with some question and get the right answer right away.
and of course it is not like that.
hehehe.
added a new feature to my software and updated the tfm yesterday.
now i’m sitting and seeing a lot of ai bots doing rtfm…