Currently Archicad AI is just a useless gimmick

[Wlad222]

The current AI feature for selecting fills in Archicad is not practical because it requires an overly precise description every time, including specifying things like foreground color pen and giving exact pen numbers. The main advantage of AI should be its flexibility and ability to understand natural speech, but right now, if one types "select red fill," the AI fails unless the full criteria are spelled out in technical detail. The regular Find command already does this well enough without AI—there’s no point in using AI if it does not make the workflow faster, easier, or more efficient.

True AI should understand and convert everyday language into the right commands automatically. Users shouldn’t have to memorize and specify all parameter names or numeric values. This rigid workflow defeats the purpose and convenience that AI is supposed to bring, and currently, it only complicates the process instead of simplifying it.

3/10 disappointed.

[Bricklyne Clarence]

---

@gpowless wrote:

.......

The problem with AI being incorporated into ARCHICAD is the fact that it too was rushed, in order to "beat" competing software from claiming it first. The problem with all AI is that its "intelligence" has to be populated and instead of taking the time it requires to fundamentally fill the datasets with intelligent meaning, they rushed it. So, what is happening is that they are using YOU - the users - to make up for their laziness. Every time you use it, you are teaching it (which is true of all AI released on the market today). All your frustration and time delay at company expense is being wasted on doing something that Graphisoft should have done before its release.....

---

Which makes it even more mind-boggling that in addition to releasing this half-baked, and not-ready-for-purpose tool, they then decided to limit the number of  people using it to Subscription-only users - and in doing so lock out many many more people who could have provided useful feedback and data for them to refine the AI's sorting algorithm.

Why would anyone want to limit the sample pool of user-data source (users) in a technology that specifically relies on getting as much information from as wide and as far as possible?

It's almost like they themselves don't understand the very technology they're trying to sell us, and to use to sell their program, nor how to actually make it better.

We should just fast forward to the part where they give up on it and move onto their next jangly shiny thing (I believe that's project Aurora - another one locked to limited users testing it. That'll definitely make it really robust and ready-for-use on release. When have they ever gone wrong with this strategy?)