It is not. It’s trained on our internal docs, objective reference material eg: open-source documentation for things like the samba package - not forums, Reddit, blogs, etc.
1 Like
You mean we can’t teach it bad words? Drats.
1 Like
I use AI with caution. I was trying to google an answer to something but I just couldn’t find the right key words. One word that I had to use was kicking google off on completely the wrong direction. I wasted 45 minutes trying to bend google to my will.
Grump, grump. Ask ChatGPT the same question and it provided me the insight that I was looking for. 45 minutes I will never see again.
Since then, I flip back and forth between google, ChatGPT and copilot as a tool to guide my research and code check my bash scripts / python / vba.
I did see a recent article about a teacher that asked her students to get chatgpt to write their essay for them … and then research why and how chatgpt got it wrong.
This is a tool. Just an updated abacus, slide rule, calculator, computer, …
1 Like
LLMs have their place. If you know one programming language and need a way to do it in another, a tool like co-pilot can be a lifesaver. While I can noodle my way around basic VBA functions pretty easily, advanced matrix operations eluded me. Co pilot to the rescue.
But you have to know enough to spot issues. Like google suggesting to me today that 3.59 inches was equivalent to 2’ 7”.
1 Like
You can’t.
2 Likes
My personal experience with LLMS:
strong textLLMs are generally very good at reorganizing things for you. I really like tables, so I’ll feed it things and have the relationship kinda shown in a more human readable way.
I have frequently posted these kinds of tables here. I think it makes summarizing troubleshooting information easier.
Giving an LLM objective or mathematical manipulation type inputs result in best possible outcomes because it’s not trying to do a “human—>machine—>humanish” subjective response.
but it does a poor job if you ask it to explain to a wider audience. It’ll kinda be right.
Sine note: Im not sure how a ban would even be enforceable.
Just detect these posts by AI, duh.
2 Likes
Not a huge deal, but it might be cool to add some API knowledge to that mix – from my experiences, it seems to know nothing about that area, e.g. the exchange below:
TrueNAS AI Search
Ask me questions about TrueNAS! For best results, include the software version and any other specific details about your topic.
What are the API calls for managing SED passwords?
I’m sorry, but the available documentation does not provide details about specific API calls for managing SED (Self-Encrypting Drive) passwords in TrueNAS. The documentation primarily describes using the web interface or command-line tools like
sedutil-clifor SED management. For Enterprise users on TrueNAS SCALE 25.04 and later, SED management is available in the UI, but API details are not specified in the provided sources.You might be able to find more information by entering a keyword in the search bar on Search | TrueNAS Documentation Hub.
Your feedback helps make this bot better! Click the thumbs-up or thumbs-down icons to rate responses. Thanks!
For the most accurate answer, please specify your TrueNAS version (e.g., 13.0, 24.04, 24.10, etc.).
What is the midclt call system.advanced.update syntax for updating the sed_passwd value?
I’m sorry, but the provided documentation does not include the specific syntax for using the
midclt call system.advanced.updatecommand to update thesed_passwdvalue in TrueNAS SCALE. The available sources do not cover this particular API call or its parameters.You might be able to find more information by entering a keyword in the search bar on Search | TrueNAS Documentation Hub.
Your feedback helps make this bot better! Click the thumbs-up or thumbs-down icons to rate responses. Thanks!
For the most accurate answer, please specify your TrueNAS version (e.g., 13.0, 24.04, 24.10, etc.).
What you’re running into looks like a training issue.
It’s why some simple tasks are readily AI solved via mimicry while others elude same because there aren’t twenty examples that may have been sucked into the AI corpus already. So, with only one solution, the AI may be programmed to bow out.
Alternatively, the AI may be having as much fun finding applicable information as I sometimes do. 
@Constantin is pretty close to the mark there. We intentionally have the AI Search instance trained to have low “creativity” and avoid inventing answers to questions that are not answered by available documentation sources. It does have access to the API docs at api.truenas.com, but you happened to ask very specific questions that it wasn’t able to answer.
The first answer it gave was based on Managing Self-Encrypting Drives (SED) | TrueNAS Documentation Hub and I think the more specific issues it had answering this particular question are (1) not all calls are migrated to the new API in 25.04 and I have a feeling some of its default training was telling it not to look ahead to 25.10 unless specifically asked, and (2) you specifically asked for sed_passwd which doesn’t exist even in the 25.10 API docs.
That said, I would like if it was a little better at figuring out that midclt suggests you’re looking for an API method and sed_passwd is pretty close to the actually existing system.advanced.sed_global_password, so we’ll see if there’s a training logic tweak we can make.
1 Like
Closing this as it is not really a feature request. Thanks for the great discussion everyone.