Using ChatGPT and the Dorico First Steps pdf's

Even if AI could train on Steinberg material, would it not have the same fallibility as human searches that pull wrong answers from information based on prior versions of the manual/program?

Since Steinberg enforces a specific model on their documentation across platforms, it would be up to Steinberg to implement AI-assisted search into their documentation when they think it reliable and productive. Even then, implementation would likely make Lillie’s job massaging meta-data even more complicated, not that she would not be up to the task (and she and the rest of the Team would still have to clean up situations in which the AI produced wrong answers.)

I don’t think AI is reliable enough yet to be depended on in this area.

1 Like

How would I condense a piece of music in Dorico 1.0


Dorico 1.0 does not have a feature to condense music directly. To reduce the amount of bars or staves, you’ll have to manually re-write the music or adjust the formatting.

However, starting from Dorico 3.0, a new feature was introduced called Condensing. This feature automatically creates a conductor’s score that is as easy to read as possible, without requiring any manual intervention.

If you’re still using Dorico 1.0, it might be worth considering upgrading to a newer version to take advantage of this and many other features.

1 Like

Looks like ChatGPT4 is somewhat limited, but what will happen when these LLM’s have realtime access to the web? Surely that day can’t be far off, if it hasn’t already arrived?

What are the improvments in Dorico 5 over Dorico 4?

Chat GPT4 replied

As of my knowledge up to February 2022, Dorico 4 is the latest version of Steinberg’s advanced music notation software. There isn’t a Dorico 5 version released yet. Any information about improvements and new features in Dorico 5 would be purely speculative. Please check the official Steinberg website or trusted software news sources for updates about new versions of Dorico.

1 Like

You can’t rely on ChatGPT . I asked a mathematical question and the answer was wrong, although it sounded very professional. So, for people , who don’t know the right answer, it can be quite misleading.


Yes, ChatGPT is much like politicians: They talk very well, it’s just the content that’s not too reliable :wink:


To be fair, they do put a disclaimer saying, in effect, “buyer beware”. If you’re interested in mathematical problems, you can try the Wolfram plugin to ChatGPT4+.

In the meantime, use those good old fashioned log tables!

Lillie - I feel the same way… 'nuff said!


It’s bad at math because it’s processing words by weighted choice. "Roses are red, violets are “____” It’s going by training on a vast data base of what follows that set of words most and assigns a weight. And it has done this millions of times assigning different weights and comes to its conclusion. (This proves the human brain is just a big piece of a lead weight. (just kidding)

That’s the way I understand it. Someone please correct me if I’m wrong.

That’s how I understand it, billions of connections. Amazingly clever stuff!

The answer: no, was to be expected if you ask me :wink:

With any simple pdf software you can cut the pdf in half, remove all images or whatever. And why not only copy the text or convert to a text file (and then save again to pdf if needed)?

Of course I tried it already, expected the PDF to be encrypted, was VERY surprised when it wasn’t, pushed it through NitroPDF, downsampled all images, got it to 38 MB.
Then I ran into the 300 page freebie limit on the site. :neutral_face:
I might go for a trial period.
It would have been fun to play with for a bit, since I love to fool around with cool tech.

P.S.: I dunno about all this. I respect Lillie immensely, and didn’t even want to post in this thread anymore, since I got scared of her… :face_with_peeking_eye:
But I can still read for myself, self-taught when I was 4… :wink:


This is an interesting thread/topic and one which we will discuss internally within Steinberg. There is undoubtedly an opportunity to look at these new technologies in making our expertly-written documentation even more accessible in a conversational format.

As mentioned already here, our documentation is copyright, so while experimenting could be seen as fair use, it should be up to us if/when it comes to deploying this kind of technology in a way that serves you as our customers to the high standards we’d expect.

I personally see this as an assistive tool to guide users to the most appropriate documentation, but only if it actually does that reliably and doesn’t “hallucinate” answers which are misleading. This is why we shouldn’t rush to adopt something like this and will consider all the implications carefully.


It also depends on user to ask the right questions. Many do, right from the start, but there are a significant number who “move the goalposts” partway through a thread when they realize they have not given sufficient details to start.

(Now when AI can combine the user question and analyze the Diagnostic Report without human intervention: that will be something.)

I wish the idea of generative AI sometimes “hallucinating” things would be not as pervasive as it has become. The reality is that LLMs and similar generators do hallucinate each and every single time; it’s just that these hallucinations, by definition of the process, are more likely than not to match up with our expectations.


If my previous experience is correct, Chatgpt does more than spit conversational replies. You can ask it to clarify, reexplain in different terms, and if you’re really lost, explain something in terms that a 7 year old would understand. Or you can ask it to explain in numbered steps in actional movements of physical steps if it has such access to information. You can ask it to give all possible ways of doing the same thing after you tell it what to do.

Like everyone has mentioned, the output can vary on accuracy, so you take it all with a grain of salt.

It might be useful to have it access all the steinberg employee responses on the forum as well as the official written help. But I don’t know how you do all that. Or if you can.

Even if they could, presumably Team responses on this forum have changed as the program has evolved. Users and the AI would have to be very particular which version they were using.

Apparently that’s why Elon Musk bought Twitter, so he can train his xAI on it - what a thought!

A good point Derek. My new (draft) strategy when trying to solve a Dorico puzzle is to first see if the forum has an answer already, then if not to use one of the LLMs to see if any could help, but if still drawing a blank, to make a plea for help on the Forum. Hopefully that might reduce the workload on the experts at Steinberg. (same goes for Cubase)

I just access the on-line manual from within Dorico. Having payed attention to Lillie’s many posts on how it has been organized, I can usually find what I need fairly quickly, and if not, I can search or post here on the forum.