Recently this forum gets flooded with inquiries on very simple tasks in Dorico. It seems a lot of new users try to relate to answers they get via AI - which are mostly not correct. As a result, real people, real “brains”, really helpful forum and fellow Dorico users have to help with “fixing mistakes” which are actually fixing wrong answers received by AI. I feel that this is, even if not intentional, mis-using the generosity of forum members to “train” the AI in the long run. Just my personal impression..
Computers can now sing, write songs, create paintings and translate poems.
The only thing they are having trouble with recently, is providing factual information… ![]()
I have started to shamelessly set AI-using forum dwellers to ‘ignore’ for want of a true block function.
Badly and unoriginally.
Of course! My point was that it used to be the other way around! If I wanted a bad painting in the past, I would have done it myself… ![]()
To which I would add the admittedly OT plea for all to consider limiting their use of AI-assisted searches for any simple information. The power consumption and water use for such searches is considerably higher. Most browsers — but not all, I know — have a simple setting to default to no AI-assist.
There are currently a huge number of topics which are basic beginner questions. I believe all of these could be answered if the users took the time to read the manual, Admittedly it is 1200 pages long, but it certainly can be studied in detail, over time, and returned to while engraving constantly. It is a superbly well written manual, quite outstanding.
I run some Discourse forums myself. My suggestion that will of course be ignored by Steinberg is to make a new Category (Dorico is just a Discourse Category) called Discourse Beginners, a place where beginners can ask their questions, and stop cluttering this forum category. Just a suggestion.
I see some newbies on Reddit asking similar questions - I’ve been telling them they should come here instead, but maybe that’s not a good idea!
Better they come here.
Jesper
The Dorico manual is not some immense, obtuse monolith.
On the contrary, I have found it to be incredibly useful and clear.
It is only when I need further information that I even ask questions on the forum; typically about extensions of the techniques learned from the manual.
That people ask questions obviously answered in the manual, or via the basic official Dorico introductory training video, is a source of annoyance.
I’d say it isn’t training the models as such. My experience with Copilot suggests that in its eagerness to give you an answer and a positive outcome, the technology guesses rather than admit that it doesn’t know the answer. Also it seems to have limited powers of inference, to say the least
Exactly.
And, to iterate, what’s needed is a Steinberg-owned and administrated AI-based help system.
Trained on the manual, first steps and available videos, and accessible from within the program. In the sense of: “Here’s how to do this, step 1, 2,3 and 4. Want me to show you?”
Help system then proceeds to actually open all the dialogue boxes needed to solve the problem… ![]()
Done.
The last thing humankind needs is machines with a fragile ego… ![]()
What’s not being considered here is how many users are successfully using AI for their queries and never need to ask on the forum at all. So it’s impossible to assess AI’s competence rate, as this forum is the place for problems, not successes.
I have used AI a handful of times for some Dorico questions. It gave me the right answer immediately. I think it can be an incredibly useful tool. Just because it’s being abused and over-hyped doesn’t mean it’s all bad, or that it will ultimately fail completely. LLM’s are probably not going away.
Yes Dan, me too, and if the advice is wrong, it’s not actually a big deal.
You’re ascribing all kinds of emotions to something that is fundamentally inhuman. On the other hand I can think of a few folks here which this does apply to.
My thoughts too, Dan. When working with Dorico I’m finding I get 95%+ of what I need from AI. That said, my questions are pretty simple - e.g. where am I going to find this setting in Dorico? If I can shave a few seconds off hunting for things, so much the better.
Side note: learning how to get the best of out of AI is a skill in itself.
Not really, it’s just a best-fit description of the reasoning implemented by its human programmers. Though, yes, some people here… ![]()
Consulting the thoughtfully organized Dorico manual is delightful (thanks @Lillie_Harris). Consulting fellow users who provide nuanced strategies based on their own journeys is likewise gratifying. Interacting with a probabilistic text-prediction model has, for me, never been more than a miserable, empty experience, regardless of accuracy.
@dan_kreider @RichardTownsend @David_Tee , do you have thoughts about LLMs’ energy and resource consumption? I’m curious because I never see fans of the technology engage with the topic.
Like you, I get my fulfillment elsewhere. ![]()
According to Gemini (Ha!)
The estimated energy consumption for an AI on Google Search query, powered by the Gemini family of models, is about 0.24 watt-hours (Wh)
This efficiency is a significant improvement. It is reported to be a factor of 33 better than the previous year’s performance. This improvement is due to infrastructure and model optimizations.
This energy usage can be compared to other digital activities:
-
AI on Google Search: 0.24 Wh per query.
-
Standard Google Search: 0.3 Wh (0.0003 kWh). This is equivalent to powering a 60W light bulb for approximately 17 seconds.
-
GPT-4: Approximately 2 Wh per response.
-
GPT-5: Approximately 18 Wh per response. This is about 75 times more energy-intensive than an AI on Google Search query.