Continuing the discussion from What makes 6v better than 2v:
No. Absolutely no AI please
2 Likes
Just another topic to ignore.
1 Like
I think not. It is a tool which people can use if they want, but to me not a valuable and personal contribution. You could say encyclopaediae or scientific papers contain similar knowledge, but those are reviewed and edited. After they are published, they remain the same. AI could give you a different answer each time, and we wonât know for sure what sources it bases its answers on. Itâs also highly impersonal. If I post from AI my post loses all personality and character. I could go on.
As an example: I would rather know of Donald that he swims to keep fit than that he can use AI to generate an answer to a question somebody else has. To me this goes against the core of what a forum is. A place to meet and exchange knowledge and experiences with people. Anyone can use AI to answer a question they might have. Please do so privately if you are so inclined. On the forum, share your own stories, experiences, wisdom, a bit of what makes you you.
6 Likes
I guess I would answer the question with a question: âwhy, what value would it add?â The forum is here for you, the members. It is a place to share experiences, get questions answered and get help with things you are struggling with etc.
One way of categorizing the informational questions is to divide them into opinions and facts. When it comes to opinions this forum probably represents the greatest aggregation of unit-years of dCS ownership anywhere, so while opinions are always opinions, they are coming from the most experienced constituency of dCS owners anywhere. The forum regulars are also able to answer many of the factual questions from information already in the public domain, and if they arenât then the dCS participants on the forum (mainly Phil, James, me) will, where we can, try to jump in and answer questions about the products and the company. Thatâs our aim at least, and we donât do that anywhere else on the internet.
So that leaves me wondering which repositories of information outside this forum would be more suitable to train an LLM from. If there are others then the idea might have merit, if not then it might only serve to dilute the quality of the information shared here.
So for those in favor, I would be really interested to hear what value you think it would add.
7 Likes
I agree with all you have said here.
There might be one exception: AI can brainstorm like almost no human. If new ideas are needed, new viewpoints, new designs, thinking way out of the box, then AI might give us a brainstorm to start with, and to be evaluated here by ourselves.
Examples, just a AI brainstorm 
3 Likes
I actually tested a local RAG (Retrieval-Augmented-Generation) on a basic LLM fed with all current dCS Model manuals, it was pretty scary how well it answered basic questions that we commonly see on the forum!!
While I havenât seen a RAG module for Discourse, an AI module already exists, so, it kind of inevitable in some form, just a question of when 
3 Likes
This makes perfect sense to me, Anup. Training the model on high quality information (hopefully the information in the manuals is âhigh qualityâ!) provides high quality answers.
My concern is that training the model on information from the across the public internet might dilute or pollute the quality of that information. Ask yourself for instance what sort of information you will find about dCS on forums like ASR or Youtube. Lots of very strong opinions, but many from people who have never listened through dCS equipment much less owned it and never had any interactions with the company. How good are these models at evaluation/criticism of sources?
3 Likes
Perhaps there could be exceptions. If a technical question hasnât been answered to a posterâs satisfaction in three days, then an AI generated reply based on web search would be permissible as long as the sources are cited.
For example, these are the sources the chatbot used in the other thread to synthesize the answer to Kishoreâs question.
1 Like
Althugh this may be reasonable it requires following a process which may be too complex for a forum when it is clear to regular readers that many may not read the rules given in the FAQs in the first place. There is the practicality of how rules that partially allow use of AI are to be interpreted or âpolicedâ. One for the moderators.
My concern is not really with the kind of straightforward operational question ( where, incidentally, a quotation from the user manual would suffice and there is no need to involve AI) but with other aspects of what AI is capable of. In any case one may lose the tangentail but important comments that only readers with real experience may make e.g.the how and why behind the question.
If I remember correctly our first encounter with AI was when a review of a dCS product was AI produced and the instruction to chatbot ( or whatever) was that it should be positive.I guess a negative review could be am alternative. AI can also produce text in the style of human authors or typical of , for example, a magazine. Of course it can also fabricate images as @Ermos demonstrates above.
Are AI text or images copyright protected? I think not but do not know particularly as the immediate future may require amendments to copyright legislation. If they are or become so then permission may need to be obtained before use here.
AI can do far more than the limited scope of the examples here. So it may require constant observation and reaction.
4 Likes
@Anupc, is there a way that we could see and use this?
This topic was discussed on the forum a while ago. As Pete points out, the landscape of AI changes rapidly, so for a small and niche topic forum such as this one, weâre hardly going to be the frontrunners in AI content detection or the moral issues surrounding its use. In the time since the last discussion here, it has moved on significantly.
When it comes to the AI tools that are becoming commonplace for customer support interactionsâŚ
This is a forum run by people and for people, and it is my opinion that sadly that seems to be something in decline in general on the internet. I do not see a need for AI tools to be used on this forum for moderation or responding to queries - we have an active presence of numerous staff members, and I would like to think that if a user wants to get in contact with us, they can, and we will respond appropriately and in a timely manner. If that isnât the case, I would much rather work to improve our own support capabilities than resort to AI tools.
Regarding AI generated contentâŚ
Where the discussion arrived at last time this topic was raised was that it would be practically impossible for a âbanâ on AI generated content to be enforced. As with most things, weâre reliant on the Community users to help decide what is appropriate or not, and to help notify us of content that does not adhere to the forum rules with the flagging feature.
I think it is permissible for such content to be posted, provided the poster makes it explicitly clear that the content is AI generated. In my opinion, users have a right to know if the content they are looking at is not from another dCS user, but generated by an AI product.
This is very true. We did edit the FAQs to include a section stating that if content generated by an AI product is posted, it should be explicitly clear that it is AI content (see Be Who You Really Are). Whether it is read by all users or not is an entirely different matter, but this is where the input of Community users is so massively valuable. We have all managed to keep this forum an authentic space, and that is primarily down to the input of the members ensuring things stay civil, constructive and in this case, legitimate.
10 Likes
Totally agree. I wouldnât want an AI bot trained on random information from the Internet answering dCS specific questions.
But thats not what RAG is. Itâs an AI model that answers to and is trained on domain specific documentation (like the dCS User Manuals for example).
Sorry Erno, no. But itâs easy to get up and running for yourself. Iâll PM you 
1 Like
I believe AI can be a valuable guide, offering directions and even uncovering insights you might not have discovered on your own.
But ultimately, itâs your ears that must decide what truly sounds right.
I think, there have been some good arguments presented why chatbot generated answers are not necessary on this forum.
Yes ai generated content is not allowed. Donald_D kindly canât use it.
1 Like
Hi Ahmad and welcome to the forum.
One thing that will help new members contribute is to read the rules for posting which you will find in the FAQs. These are found in the hamburger menu ( three lines) to the right at the top of the page.
You need to see " Be Who You Really Are". You will find that AI generated content is currently allowed so long as it is explicitly identified:
he use of content created by machine learning / AI platforms (including but not limited to chat-bots like ChatGPT, or image generators like Midjourney) is only permissible if it is made explicitly clear in the post that content from such a platform is being posted, and exactly which parts of the post are from the machine
Should this remain given that it dates from the early days of AI which has developed subsequently ( and which will continue to develop)?
It is also worth reading the thread so far. If you do you will find a post from James ( a moderator and dCS staff member) explaining this and some of the thoughts behind it.
Hereâs a taste of whatâs possible right now
Dropbox audio clip (just under 20mins) of a completely generative AI conversation/podcast between two AI synthesized voices, built purely from RAG training on the dCS Bartok, Rossini and Vivaldi User Manuals.
Again, to be clear, these are not real people, itâs completely AI generated (in under 10mins from just the Manuals) - https://www.dropbox.com/scl/fi/n6ud8num0408fzo53ylx7/dCS-AI-Conversation.wav?rlkey=ln7qmhjoldiu5gus54tmo26wx&st=2v4upvg0&dl=0
What is meant by âOn the DC units like Bartok or Lina you get dedicated headphone outputs right on the frontâ around 4:00? Thatâs not correct anyway, the Lina comes with a headphone amp.
BTW, one doesnât need to pre-train the reasoning models with the dCS manuals to have oneâs questions answered. With web search these models are de facto âRAG-enabledâ and pull information from the internet to enhance the answer/analysis. In this case, I linked the manuals of the Lina and the Rossini. I wonât show the answer but itâs impressively precise.
The text below the question shows the thinking process of the model before it generates the answer. Itâs in German since Iâm based in Germany, but ChatGPT has standing instructions to answer in the language that the question was posed. It reads:
<<The user wants to know what the âClockâ does in relation to the Lina DAC and Rossini DAC. The manuals are probably available as PDFs, so I need to open these to find relevant information. Since the request is in English, I will also compose the response in English. I will review these pages to understand how the Clock functions within these audio devices and then provide a precise explanation.>>
