did every social media adopt a common AI/ML algo for timeline recommendation or what.

EVERY goddamn platform now recommends posts heavily biased towards the last 5-10 accounts/topics you've interacted with.

I have to manually go delete my history/"saved data" just to get a fresh air of posts every few months.

I think it's been that way for a long time. Things are just more on the spotlight now

I think u need to learn what ML and AI actually are.

Suggestions based off your last interactions is often just a simple af algorithm.

That said, are you totally unaware of the fact that both google and siri have been not only using your cookies, history, anything you type/read on your devs, ever... to suggest like everything? Even what shows up in your searches, play stores, search engines, the url bar suggestions of what you might be typing... same w/your keyboard, etc??

Personally, i know waaay too much to ignore that crap even though i know that itd likely never effect me directly... but with my history and skillset i dont even allow the '...is typing...' things and beyond to track me/anyone on my property, so ive made methods around it for over a decade.

I had to allow google unobfuscated access for about 30m several months ago. Talked to/about an employee in Myanmar, was writing py. 1d later open e-book app (very rare use). It heaviliy suggests a book, "Burmese Pythons"

Definitely been like that on YouTube for a long time. I have to be very cognizant of what I view if I don’t have all my future recommendations to be filled with it. Pretty annoying.

@awesomeest Ive worked in AdTech (not in AI dept but still) previously so im aware how unethical the companies are and to what extent they go to, to track users' whereabouts

But at least the AI before was a mix between fresh and what one's used to. It wasnt choking users

This new time-bound-isolated recommendations is a recent phenomenon from a user's PoV

@azuredivay can you explain what you think AI is? I dont just mean "Artificial Intelligence", i mean what that means/is in computing terms.

@awesomeest multi-variable pattern recognition would be the gist of my understanding of it,
Of these multiple variables, which ones matter more/less eventually lead to a weighted graph we query against

@azuredivay technically it's anything that "learns", as intelligence is specifically an ability to learn.

I've had this discussion several times, including with several gifted/accomplished devs in the field, over ~15yrs. Basically, no matter how complex a pattern or hardcoded function is, that's not learning. If it's responding, and processing, off of it's programmed ability directly, it's not AI. The mechanisms and theory behind it all gets quite acutely semantic and verbose. An oversimplified version would be something akin to taking base operations and adapting to a significantly higher layer, even using the newly created logic as a basis for evaluating future operations. It doesnt need to be super complex, just needs to learn. Technically i wrote an AI on to an arduino back in the late 00s, and that was just to make a, hard to detect, beep come from my older brother's closet he wouldn't clean.

@azuredivay you've described the Eliza family of pseudo-AIs, a 60s-ish thing

current AI's take no variables, their "parameters" stands for data matrix dimensionality

----

Eliza => https://github.com/emacs-mirror/...

@awesomeest but any "learning" is just a change of data or addition to multi-variable weighted graph no? unless theres a new way of defining/storing "learning"
@kobenz i didnt mean predefinied variables, but "take in data, figure out which diversions are its dimentions, then weigh them differently over cycles", i.e. 'multi variable weighted graph'

Either ways though, my point simply was that AI is being heavily biased towards recent changes, and this behaviour is more apparent than before, in a bad way

I'm actually somewhat concerned about the recommendation algorithms in social media too. It's important to note that each platform has its own methods and approaches to recommending content, including the use of artificial intelligence and machine learning... According to this social study https://papersowl.com/examples/... It emphasizes that recommendations on social media have become a topic of discussion due to growing user frustration with algorithms that can exacerbate fragmentation and narrow perspectives on information... I'm worried.

@azuredivay learning isnt simply "adding data".technically it is adding data... literally everything on a computer is, in fact, data.

AI, when it's "learning" needs to be written in such a way that it dynamically adapts it's own code base.

Think about it like this over simplification:
If a little kid has never interacted with any animals before, just knows flash cards and the names like dog and cat, and woof/meow from their parents saying them. If that kid wanders off and finds a cat, and being curious, corners it to go pick it up... if that cat suddenly looks straight at them, makes a defensive stance, shows teeth and a loud hiss (nothing like "meow") the kid may still try to pick up the cat... and get scratched to hell.

Same kid then encounters a dog the 1st time, approaches it, with or w/o potentially learned caution, when that dog makes the defensive stance, etc, despite it being a different animal and a different noise, it's unlikely they will keep approaching.

^ learning