Anybody with a newer iPad notice the mail (possibility) change?

[omega man]

I just updated to iOS 18.4.1 and it has the option of categorizing emails into 3 stock categories along with the ability to add a category you make yourself.
The categories seem pretty accurate, kinda amazing.
Not sure if I am going to leave it enabled. Any thoughts?
OM

 

[r0ckrat]

Gmail introduced something similar a few months ago, and I've really been liking the ability to (and having it automagically recognize) the different "buckets".

 

[GTRider]

Jury still out.
The categorization is still a bit flaky—emails shifting from one category to another after a few days, or appearing in multiple categories. My chief disappointment with the update is that it still doesn’t address some long-standing issues, like previously sent emails reappearing days or weeks later in the Drafts folder.

The unfortunate reality is that the ongoing shift from email to messaging is likely throttling development on OS-provided and stand-alone 3rd-party email programs in both the Apple and Windows worlds. Probably because it seems easier and more palatable to the public to slide advertising into messaging apps than email apps. Guess that’s why my main email app is still one that dates to the 90s—but the clock is ticking there, too, as the developer is contemplating retirement and could close up shop…

Best,
DeVern

 

[zer0netgain]

I turned that nonsense off. I don’t need an OS deciding what mail is a priority for me.

 

[kbasa]

I turned that nonsense off. I don’t need an OS deciding what mail is a priority for me.

I use the Google tools and they're pretty effective. I'll try the Apple stuff. I get boatloads of email and most of it is garbage. The Google tools do a decent job of moving the junk into folders and out of my way, leaving my inbox almost all stuff I care about. it's really good at managing the mail I get from some mailing lists I'm on.

Nonsense? I dunno. I'm always good with computers getting rid of garbage for me.

 

[omega man]

As I mess around with this new feature, when you (I) receive a mailing from one of your “regulars”, clicking on the newest email finds all the correspondence, new and previous, available for viewing.
Might be a good, or at least a different search method.
OM

 

[mikegalbicka]

AI at work

 

[kbasa]

Gmail introduced something similar a few months ago, and I've really been liking the ability to (and having it automagically recognize) the different "buckets".

I've had it for a couple years, anyway, and use it for both my personal and work accounts. At work, it's super handy to filter all the distribution list stuff I get.

It’s better than the endless Outlook rules I used to have to build. I don’t think it’s true LLM AI, but more likely a CNN system built on Google’s BERT technology.

 

[PGlaves]

I've had it for a couple years, anyway, and use it for both my personal and work accounts. At work, it's super handy to filter all the distribution list stuff I get.

It’s better than the endless Outlook rules I used to have to build. I don’t think it’s true LLM AI, but more likely a CNN system built on Google’s BERT technology.

Do you mean Bert, or is it Ernie??

 

[kbasa]

Since you asked: https://en.wikipedia.org/wiki/BERT_(language_model)

This is probably more than you wanted to know, but basically, BERT is a gizmo that can predict what words come next in a sentence and what sentence comes next in a paragraph, a skill it uses to transform or analyze text.

They're probably using BERT to scan the text and a categorizer tool to put them in folders. I did some reading and it appears that if you start to help it organize, it'll learn from your actions and update it's categorizing model.

 

[omega man]

Since you asked: https://en.wikipedia.org/wiki/BERT_(language_model)

This is probably more than you wanted to know, but basically, BERT is a gizmo that can predict what words come next in a sentence and what sentence comes next in a paragraph, a skill it uses to transform or analyze text.

They're probably using BERT to scan the text and a categorizer tool to put them in folders. I did some reading and it appears that if you start to help it organize, it'll learn from your actions and update it's categorizing model.

OM

 

[PGlaves]

OK, I get it. It makes total sense. I understood every word except:

"BERT is trained by masked token prediction and next sentence prediction. As a result of this training process, BERT learns contextual, latent representations of tokens in their context, similar to ELMo and GPT-2.[4] It found applications for many natural language processing tasks, such as coreference resolution and polysemy resolution.[5] It is an evolutionary step over ELMo, and spawned the study of "BERTology", which attempts to interpret what is learned by BERT.[3]"

 

[kbasa]

OK, I get it. It makes total sense. I understood every word except:

"BERT is trained by masked token prediction and next sentence prediction. As a result of this training process, BERT learns contextual, latent representations of tokens in their context, similar to ELMo and GPT-2.[4] It found applications for many natural language processing tasks, such as coreference resolution and polysemy resolution.[5] It is an evolutionary step over ELMo, and spawned the study of "BERTology", which attempts to interpret what is learned by BERT.[3]"

This is kinda long, but if you're truly curious, I hope it's helpful.

BERT is trained by exposing it to tons and tons of language in books and other publications. Tokens are basically words. For context, "latent representations" means that it is looking at both the words and their context as it works to "understand" how words are used. I used to work with a tool that performed "latent semantic indexing" which meant that I could show it text and it would show me similar text. Or I could write up a statement and it would go find similar text in an unstructured data collection.

In this paragraph, they discussing how it's trained and how it learns. Basically, after the tool has been trained on what English looks like, the tool converts words to "tokens", then removes a couple tokens from a sentence. As BERT tries to figure out what word goes in there, based on its training, it checks itself after each attempted substitution. It compares its substitution against the sentence in its original form to see if it correctly inserted the appropriate term. It continues to do this until it gets to a high degree of certainty that it can form the sentence correctly and is writing coherent text.

Polysemy is basically a way to figure out which specific meaning of a term is appropriate. If I say "I went sailing by you when i passed you", the system will try to determine whether "sailing" means I was in a boat or it was a different usage of that same word. "coreference resolution" is a similar type of process to resolve a term to it's distinct meaning in context.

For context, "latent representations" means that it is looking at both the words and their context as it works to "understand" how words are used.

This is a really core part of any AI system. It has to be trained via exposure to hundreds of thousands or millions of examples. This is also why public AI tools are such garbage. They've been trained on unvetted trash on the internet, so they'll generate nonsensical or untrue results. But in a commercial setting, where the models are fed highly vetted and accurate information, AI can lend huge benefit.

I'd expect that Google is using a form of BERT behind the scenes to funnel email into specific containers in your gmail inbox. I'd expect that it's also being used to support Gmails pretty decent spam filters.

BERT is a pretty simple form of AI, but a very, very powerful one.

I hope that's helpful.

 

[PGlaves]

Thanks

 

[mikegalbicka]

Gmail has always had the best spam filters compared to any other service I have used, which is most of them, IMHO.

 

[omega man]

The speed in which this new iPad can switch/rearrange the mail categories is the blink of an eye. Double tapping on selected categories reveals more options. Interesting.
OM