Hi friends,
We’re currently improving the Aqara Forum Bot and would love to hear your suggestions.
If you’ve ever noticed inaccurate replies or have any ideas on how the bot could be more helpful, feel free to share them in the comments with as much detail as possible.
Let us know how we can make the bot better for you! 
At all the bot can be helpful- but the bot gets really annoying when it spams all the time in an unacceptable amount.
Also double or unasked answers should be reduced
I think it need a more comprehensive knowledge base of the products, how they work and the technical aspects of their use. It’s current knowledge doesn’t seem to go far beyond the standard product page information.
I had 116 unread messages from the robot, dismissed them all. Should choose whether to have the robot to reply when you creating a topic or replying to one.
Never answered a question I’ve asked with any relevant answer. Waste of time and just adds annoyance. Bin it off!!
I have noticed that the bot automatically adds information in reply to posts about devices that have received a firmware update. This is basically a useful feature.
The problem with this, however, is that very old posts are displayed as new or unread, and new posts are actually lost in this flood.
Therefore, I would argue that it would make sense for the bot to only respond to posts that have had no activity for no more than three days. And only then if the OP has not received any replies in the last two days.
Alternatively, the bot’s responses could be configured so that they do not notify all subscribers or update the latest activity. For example, it would make sense to only notify the original poster (OP). However, I do not know if the forum software or plugin offers this functionality. Probably not.
Similarly, it doesn’t make sense for the bot to post multiple messages quickly when nobody else has posted anything. In this instance, a prohibition for a period of forty-eight hours would be suitable. I wonder why the bot doesn’t post all the information at once, rather than creating one meaningless post followed by one that is somewhat meaningful.
Make the bot, learn especially about the forum experts, It needs to learn what they can and can not reply to, I dont mind you guys sharing with the bot what devices I have or have had with the bot so it knows exactly when to mention me.
I strongly suspect that it works with RAG (Retrieval Augmented Generation). It is most likely the Discourse AI bot.
The bot cannot learn who the experts are. It only finds and uses existing posts by matching similar topics through the vector database. While it reuses content that has already been posted to the forum, uploaded as a document, or sourced from other external sources, it doesn’t train or change itself.
Technically, each post is stored as a high-dimensional vector, and the database calculates the distance between vectors to find similar posts, which are then passed on to an LLM. With a bit of luck, the LLM will receive the question and the answer at the same time if similar questions have already been answered here in the forum. Or, for example, if the instructions for a device have been uploaded as a document. The LLM can then formulate its own response using all the information it has received.
If you were to write a post about the experts and the devices they use, it would be relatively unlikely that this post would always be forwarded to the LLM. In addition, LLMs are limited in terms of the maximum number of tokens they can process. With ChatGPT, I believe this is around 4,000 tokens, which corresponds to approximately 3,000 words. However, this is different from AI and models. Therefore, the information that can be passed on with the help of RAG is always limited. In addition, there is usually a prompt that contains further instructions for the LLM. To ensure that the LLM always knows all the experts and their devices, the information would have to be written in this prompt. But someone has to maintain the prompt, and as I said, you can’t write too much in the prompt, otherwise the LLM will forget the beginning. And each additional token also costs more resources (money and electricity).
The bot is completely useless and should be deleted.
In general I appreciate the bot and can in some cases see it’s providing useful help to some of our newer members of the community which is great.
If we then look at the many commentary posts it makes that is being a bit much in my opinion, but the part where it sometimes add FW update information is a great touch.
As some of the experts have commented too is that it would be beneficial if it had deeper learning of what devices different experts or super users have also in order for those to better help the rest of the community, but looks like that would require a completely new bot set-up 
Como se pueden traducir los comentarios del forum al español?
Gracias
Haga clic en el enlace y seleccione español. https://forum.aqara.com/u/luisgomezbus/preferences/interface
@AqaraOfficial One big thing from my side:
Please make the bot stop automatically tagging other users — experts, creators, whoever. It’s not helpful at all, especially when the question is about something nobody can fix (like products not being available in a certain country).
My inbox is getting absolutely flooded, and I’ve already stopped checking it. That means I might even miss messages where I actually could help.
So yes… less tagging, more signal, less noise.
Always remember: “If your car is smarter than your home… you’re driving into the future while living in the past” 
Check this out: Topics to remember – Bookmark and keep coming back!
I concur. I especially don’t need the BOT telling me I’ve done a great post. It’s meaningless unless a human says it.
I often find the bot’s unsolicited contributions annoying. It should only respond when it is mentioned by its user name. The quality of the information provided is also often very poor.
Please shut down the Aqara Forum Bot!
I am not in a forum to talk to bots.
If I have question, which an AI can answer, I will ask the AI directly.
