@dfay
1.) Regarding the rendering, what macOS version are you running?
2.) I hardly ever use the AI to auto-classify research articles.
There are many posts over on the DTPO forum, explaining the process and underlying structures best suited to having the AI work effectively, but given it is essentially running algorithms that are based on content, one would need very diverse ‘areas’ of research, to have it work effectively in all scenarios.
The – justifiable, imo – refrain from some long-time DTPO users is that the AI cannot replace your thinking/insight into the content of the data, it is there as an Assistant.
I agree with this, but as is apparent from my points below, there certainly are scenarios where it works so well, to the point that it does do the thinking for me, in sorting specific types of data.
Simple example - click to open:
But by way of (silly) example – and purely for new users to DTPO who might not be familiar with what you raised:
Imagine a sports-journalist A that covers F1, NHL, Football/soccer and Curling.
This person is going to see far better AI/Auto-classify results, where he/she has different folders and sub-folders that cover those distinct sports.
Whereas sports-journalist B that writes solely on football/soccer, and has groups focusing on the different international leagues/World Cups/Fifa regulations/Rules/Major players/Games etc., is probably going to have less accurate results – given how a single article could talk about Messi in the Fifa SWC, his career at Barcelona, playing for Argentina, and why he thinks rules X/Y/Z, need to change.
That’s not necessarily a given. But it can pan out that way.
As can the scenario where journalist A runs into similar problems to journalist B, but in the context of the individual sport types he/she is covering.
But depending on how journalist B’s folders are sorted – and how many other articles are already in those folders for the algorithm to learn from – I think it is understandable that the AI is not necessarily going to know whether to file that article on Messi under the Spanish League, World Cup, Rules or Major Players folders.
However, if journalist A had folders in their NHL group, where each week an Excel spreadsheet containing that week’s league scores were placed. Or had these, but based on years – over time, again depending on how much is already in there – the AI will work out where that new Excel spreadsheet on the NHLs scores needs to go.
Similarly, if journalist B always downloaded a particular reporter’s review on games played in the Spanish League, over time, the AI would be able to auto-classify these as well.
Turning to something more concrete, I ran into this problem constantly in the beginning, since I had 4 major comparative areas, all in one DB.
So Article A would be talking about what is happening in Country’s 1, 2, 3 and 4…
Added to this, similar terms are being used in all for comparative areas – so that doesn’t prove a point of distinction either.
I then split that DB into 4.
But that still doesn’t necessarily solve the Messi example (collapsed) above – again, depending on one’s own data.
But in spite of this, I still see the AI as being very useful in specific examples – where I am more interested in filing repeatable instances of a type of file, or where I have those top-level broad groups, that I just want to get things into initially – as filed from the top-Inbox – before I get more granular after the fact.
Put differently, I view the Auto-classify feature as something that works very well in specific contexts, but is not something that will work in all contexts, equally well, if at all.
That said, the See Also/Magic Hat function is something I use frequently, even if just to quickly cycle through 2/3 articles about a specific point.
In other words, I know Author D, G and Y wrote about it, and I could search for/navigate to each – but find that when reading that article by D, the See Also prompts me with the articles by G & Y in any event, so I often use it instead of searching/navigating, since its quicker.
I’m not aware (but would love to hear of examples) of any consumer products that have an algorithm that will be able to magically sort text-heavy journal articles, in a manner that the average/bulk of users would find useful.
But I am from the liberal arts/social sciences side of the World – so that’s simply my perspective, given how I know the articles to be written in my context.
And I’m (obviously) not sure what your field is – and maybe, given what it is, and what you fed into DP – the AI should have been able to do the necessary, and didn’t.
But there are many factors at play – so what doesn’t work at first, can very well start working over time, as the DB grows and more structure becomes apparent to the AI.
At least that has been my experience of it.
