Maybe a bit dramatic (LLM generated ) but I hope some food for thought:
Lately I found myself using Gemini, Claude and the lot to, quite successfully, solve problems in my Home Assistant instance I was unable to do in the past.
In doing so, I realised a lot of you are probably doing the same. I think this holds a risk. We train big AI to become even bigger. If we don’t leave a trace, the forums will shrink and new large language models (Why not a HA community LLM) will lack a lot of valuable resources to train on.
It’s a small effort to ask an LLM to make a step-by-step procedure of your solution and post it in a relevant chat.
And if AI is currently doing it in a bad way, by doing so you will reinforce the AI in its bad choices. If a lie is repeated often enough, it will be seen as the truth...
But you need not worry. This forum and all other kinds of content is filling with AI slop faster than you can imagine.
This is a very interesting topic. I have to confess I have used some AI to speed up writing ALOT of my code lately, but I test it thoroughly and also have found some bugs which I have fixed, but it did make me MUCH more productive. Here are just three of many examples:
A huge amount of code tying together sensors from different vendors (enriched by daemons which create derived "vetted" sensors enriched with additional metadata) to create a full blown alarm system using a zwave-connected ring keypad for the front door with HA and Alarmo, as well as twilio for standalone voice and text messaging as well as being tied into noonlight as a 24/7 monitoring service that calls the authorities if needed - so when authorities are called they actually get enriched information such as "southeast living room window opened" etc... (I am texted first, if I do not reply then my cell is called, only then if I do not reply the call goes out to police)...
also to write daemons on multiple RPI's collecting data from SNMP and Omada API calls to gather every possible piece of information from my large home network with gazillions of devices to fully populate an influx DB in near real-time which drives numerous grafana dashboards measuring every single aspect of my network which allowed me to tweak it to perfection and improve it's reliability - every single IOT device tracked and connected as strongly as possible to the network...
I have an AmbientWeather WS-2902C weather station feeding into into Weewx and HA. Handy to remind me to close specific open windows when it starts reaining etc. AI helkped me improve daemons I had written which feed data into my HA instance from my weewx setup, for a dashboard showing the status of the rest API calls made to numerous weather websites all over the world from my weewx instance (for ridiculous things such as this weather website in germany showing my weather station in NJ)!
So:
A. There are rules in the forums here to not just paste AI generated stuff - the bad stuff which I understand might be posted is also called "slop"... what EXACTLY are the rules - does someone have a link where we can read them exactly as written?
B. I am hyper about security due to someone tring to steal my identity once (no Alexa in our home, none of our cell phones have 'wake up' words enabled), I refuse to implement any code which actually relies upon any calls to AI at runtime (when it is running). AI is used ONLY to create code and I implement said code ONLY when I have read through, understood all of it, and test it. So I have NOTHING relying upon any AI tied into my systems, all code I write including vibe coded is cut and pasted from the chat window, I would STRONGLY suggest everyone using AI for code writing do the same. For example, I noticed that the Claude app/integration whatever it is for HA - gets access to all your files, and this includes "!secrets.yaml" I CANNOT BELIEVE that people are ok with that!???? Same problem with Codex I believe (which I think is now being folded back into ChatGPT)....
C. Due to the above and many other concerns, I am very interested in where this all leads... I am not worried about forums would disapear, I think just like people thought computers would make our society paperless it actually created more paper - so AI will cause MORE activity on forums from people.... thoughts anyone?
The working description that has been posted previously by multiple Mods and Admin is essentially:
Don't post AI-generated content as an answer to a question or request for assistance.
If you create a post sharing a custom integration, blueprint, or other project that includes AI-generated code or content, you need to disclose that in the first post.
Use of LLMs for translation or writing/grammar editing is allowed. But, the preference is that users disclose that as well. This is to avoid the appearance of plagiarism, which is prohibited by the forum guidelines.
) but I hope some food for thought: