Comment by at_a_remove | Hacker News Reader

Comment by at_a_remove | Hacker News Reader

Parent Full thread

Ah, that word "just." Whenever I see "just" in some kind of kind of technical demand in the first line, I immediately know that there's a lot of stuff to unpack.

Here, you have requested something fairly close to a general AI which understands both written language, images, and even more fun, images of language. Cartoon violence right on down to "Will no one rid me of this turbulent priest?" And will quoting Henry II of England set off your detector?

Let's unpack your COVID post thing -- note that WHO guidelines have changed a few times. Recall that era wherein they said that there was no evidence of human-to-human transmission. Also, don't buy masks. So you'd have to factor in a time domain as the guidelines change, in addition to your generalized AI as above.

In IT, I have frequently heard requests for "why can't we just get the things we know we will like?" and such. It all boils down to that mind-reading, future-predicting generalized AI.

Why don't we just turn it on?

Okay, but companies like Google, FB, and Twitter are already putting great efforts towards implementing that type of thing in production.

Whats so difficult about letting a user toggle those existing filters on and off? Or allowing users more control over their news feed, such as allowing it to be chronological? Or hiding items based on keywords?

Its our job as technologists to not only tell people if/when their request is technologically impractical/impossible, but more importantly, to make informed recommendations of what is technically feasible and most closely matches their intended goals (which they might not yet even fully understand themselves).

banads says>"Whats so difficult about letting a user toggle those existing filters on and off? Or allowing users more control over their news feed, such as allowing it to be chronological? Or hiding items based on keywords?"<

This! And...

allow 3rd-party individuals/companies to create user-filters that users could load up at their preference. Allow a proliferation of filters to compete.

So, say, if you don't want profanity you could view FaceBook with the, say, "FaceBook-no-profanity" filter active. Or if you don't want antisemitic remarks you can load up FaceBook with, say, the no-antisemitic filter. Filters could be loaded, unloaded, switched on and off at the users desire.

So everyone could see the Internet they want or, at any time, view everything or part of it. And if you want to see what a friend is talking about, you could load the filter(s) (s)he is using. Sets of filters could be grouped, e.g., TomT's filters, Mary's filters, Mom's filters, etc., so you can, so to speak, "see the (Internet) world through their eyes".

I see where this is going...in the long run the Internet will turn into a vast wasteland with gardens of little consequence and minimal membership. It will become defunct and all else will be simple P2P over TCP/IP.

Eh, I haven't seen it in production.

I hate seeing politics of any stripe in my news feed, for example. I get enough of it elsewhere. But how can Facebook take an image, recognize that it is a cartoon, and that the cartoon is based on something political? We're not there yet. We're not anywhere near there yet.

You and I, as humans in this time period, will likely recognize a drawing of a man with kind of mocha-y skin, ears that stick straight out, large-ish front teeth, and kind of a squint as a standard issue caricature of a particular former President of the USA. We can immediately go "aha, politics!"

So I suppose we could try to train some AIs on a historical set of caricatures of politically-relevant figures. Of course, caricaturists and cartoonists of all stripes are called upon to produce new stylized exaggerations of people as they step into the limelight (or are dragged into it), so our training set must be updated on a daily basis. Hrm, some more work there ...

But then it gets into "politics you would like to hear" versus "politics you would NOT like to hear," and that is just going to be something that will require a mental model of what you do and do not like.

Right now, Amazon and Netflix cannot get my recommendations even CLOSE and their data is much better curated than what someone might type into a Facebook text box.

Chronological? Reasonable! Probably doable. Keywords? Who is putting those in? Who is curating that set of keywords?

They already have processes in place to flag and remove covid information and violent content, so you obviously don't need general AI.

Yes thank you. My examples were explicitly things that they already do in production. They were not random examples.

Ah that word 'ah'. Whenever I see 'ah' at the start of a post, I expect condescension to follow.

In IT I infrequently hear anyone say IT.

Couldn't agree more. Also the implication that users will manage or update settings to their preferences is incredible in its own right.

So, let's just do it.

Reply on news.ycombinator.com