Dolby Cinema Virtual Assistant

[Leo Enticknap]

On election day last November I asked Siri for navigation directions to the polling place.

Leo: "Hey Siri, please navigate me to the Chinese Seventh-Day Adventist Church on Mountain View Avenue, Loma Linda."

Siri: "I'm sorry, I can't find any Fountain Poo Avenue in Loma Linda."

As for my inaugural query to Dolby's chatbot, a friendly human answered the email that followed quickly and helpfully, as they always do (you need the audio processor license or else AES67 output is totally disabled).

[Randy Stankey]

Use telegraphic language when speaking to a voice recognition system.

”Siri… search Google… church… Loma Linda, California… “

Using prepositions, modifiers and other extra words will only confuse the algorithm. It was that way when I worked on VR systems, thirty years ago but it hasn’t changed much, since.

Its also better to make non-specific queries then select from a longer list of possibilities than to make the program work harder, parsing long sentences.

[Marcel Birgelen]

As for my inaugural query to Dolby's chatbot, a friendly human answered the email that followed quickly and helpfully, as they always do (you need the audio processor license or else AES67 output is totally disabled).

I took the liberty to put your question through our own chatbot, which has been trained with some of the materials freely available and some of the manuals I've found. The only thing I changed is to add Dolby in front of IMS3000, so it knows you're referring to a Dolby product, as this thing isn't Dolby specific. Maybe the answer isn't entirely perfect, but far better than what the Dolby chatbot achieved. Mind you: I did not yet feed it any IMS3000 documentation, so this comes from what Dolby provides freely on their site.

Q: Can an Dolby IMS3000 output the regular 7.1 channels in AES67 without the audio processor or Atmos licenses (i.e. just in server mode)?

The Dolby IMS3000 is a cinema server that can be used to play back audio in a variety of different formats, including 7.1 surround sound. The IMS3000 supports the AES67 standard, which allows it to transmit audio over an Ethernet network using standard AES3 connections.
In order to output 7.1 channels in AES67, you will need to connect the IMS3000 to a device that is capable of processing the audio, such as a Dolby Atmos processor or a digital signal processor (DSP).
In order to use the Atmos mode, you will need to purchase a Dolby Atmos license and install the necessary software on your IMS3000. This will allow the IMS3000 to process the audio and output it through the Atmos speakers. If you want to output audio in AES67, you need a license for the audio processor, you will not be able to use the Atmos mode.

I've underlined the wonky part of the output. This output is probably due to inconsistent or incomplete information available on-line.

This is based on a modern LLM, much like ChatGPT 4. As you can see, those models walk all over Siri and those silly chatbots.

I do think this kind of chatbot can be useful for the community, especially if we feed it with specific knowledge of the field. It will NOT beat a seasoned tech in insight and experience, but it will remember stuff you will not and it beats searching trough 300+ pages of manuals and specs to find that one sentence or number you're looking for.

Edit: Apparently, colors aren't supported.

[Bruce Cloutier]

Just like everything else on the Internet, when it replies how do you know that you should believe it? I think even 60 minutes made a big deal about how ChatGPT can lie... or at least make up stories. When real people (i.e. humans) don't know the answer at least some will admit it (without acting like they didn't understand you). There is always a chance (with real people) that from the conversation and the tone you can tell how confident or not they are in what they are telling you.

Leo, did Siri have an ulterior motive in apparently not being able to help you find your way to the polling place?

And, you know, it is still much better to physically hit pause on the remote than to yell out loud "Hey Roku, pause!" and then wait for it to flash "processing" and then tell you it is pausing before it does and you know it is safe to run to the... um beer fridge. Meanwhile the thing is listening to everything else you might be offering in confidence to your friends.

AI is still just a fascination. It is not ready for prime time. It might help some but it certainly isn't worrying itself about wasting everyone else's time. Most everyone that has mentioned ChatGPT see it as an adventure game and they seem to have had the time to try to challenge it (or break it). Then you only hear about certain more interesting responses. My mother used to apparently always win at the casino. You know that she didn't but she never bothered to mention those days.

We probably need products that are documented well enough or that are otherwise simple enough to not lead everyone to call customer service. But then the product may be perfect and yet you have to call about the billing.

[Marcel Birgelen]

Just like everything else on the Internet, when it replies how do you know that you should believe it? I think even 60 minutes made a big deal about how ChatGPT can lie... or at least make up stories. When real people (i.e. humans) don't know the answer at least some will admit it (without acting like they didn't understand you). There is always a chance (with real people) that from the conversation and the tone you can tell how confident or not they are in what they are telling you.

Those LLMs are trained on curated material from the Internet and in the end those models are only as good as the material they've been provided with. One thing where those models still fall flat is extensive reasoning, that's why this also isn't anything close to real AGI yet.

Still, when it comes down to just basic knowledge, you'll see that most of those advanced LLMs will outperform the average human on accuracy by a wide margin on almost any topic.

Keep in mind that what we've seen is only the beginning and the geenie is now out of the bottle. It's still unclear how this will work out for us, but the release of the 2017 paper "Attention is all you need" and the subsequent leak of a first, workable Large Language to the public by Facebook's AI research team may prove to be a watershed moment in human history.

[Bruce Cloutier]

My bet is that most of us do not appreciate being the guinea pigs for this when we have finally given in to fighting the phone menus and sitting endlessly on hold seeking professional and knowledgeable assistance.

How about, while you are knowingly on hold waiting for the real person, the AI works (plays?) with you (more socially) to keep you company and gather information, maybe suggesting a solution? When the real person is available he/she joins the conversation and the AI gives them a summary of what it has found out from you so you need not start all over? The customer support team works there and I bet they won't want to talk to the AI. Well, they do and probably (hehe) try to break it. I hate it when the phone system asks for your account number and then when the representative gets on the line the first thing they do is ask for your account number. Geez.

[David Ferguson]

Like the majority here, I also really hate web chatbots.

I can see that they may have uses, in some applications. The majority (including the ones I have worked on), are not smart, and just look for keywords to recommend a pre-stored database of FAQ answers. For some users, who don't bother searching documentation, this may be great. For me, not at all. If I'm trying to contact support, it's because I've exhausted other avenues, including reading documentation, trying myself, etc. so what I need at that point is to talk to a knowledgable human who can actually listen to the problem I'm having, and knows the product well enough to actually help me. Not just spit out links to the website, which I've already seen.

[Marcel Birgelen]

My bet is that most of us do not appreciate being the guinea pigs for this when we have finally given in to fighting the phone menus and sitting endlessly on hold seeking professional and knowledgeable assistance.

The thing is, that's all primitive garbage compared to what's to come. Not the same ballpark, not the same league, a different sport entirely.

Those phone menus and those "intelligent" chatbots and voice systems of the current generation are just mind-numbingly dumb and just customer-hostile. I remember just recently being on the phone with one of those systems (UPS, I'm looking at you) for more than 30 minutes, trying to get a human on-the-line. The entire thing was borderline slapstick. Eventually, the human I got on the line was about as useless as the thing that came before, but I digress.

When it comes down to modern, complex language models, those things answer stuff already much more like a normal human being than your average first-line service tech. I'm now using a ChatGPT-alike virtual assistant on a daily basis... It even beats Google on many subjects. You just handle it a question and 9 times out of 10 it just comes up with a straight answer, no more clicking through several pages involved...

[Dave Macaulay]

Today I asked what DSS200 version uses a certain CAT862 version (I have one that won't boot to tell me what version it's on, and want to reinstall the software). It gave ma a KB article with exactly what I wanted.