jack: (Default)
From: [personal profile] jack
Yeah, I thought something pretty similar, although they are a lot more specific about it.

I'm not sure if there's two specific levels, or just a big hierarchy. Like, tracking from sentence to sentence is harder than just skimming. But lots of arguments need you to track between paragraphs, and sometimes people do things like mathematical proofs I'm not sure I'd have long enough to grasp in my lifetime. On the other hand, I don't know if the same techniques programs like that already use could be extended to those larger correspondences.

I don't have much hope for "people will be more careful" though :(
jack: (Default)
From: [personal profile] jack
Huh. Yeah. Slatestarcodex also compared the program to a human dreaming, i.e. all correlations no oversight, which seemed apt.
marahmarie: (M In M Forever) (Default)
From: [personal profile] marahmarie
The thought of which (ironically enough, I suppose) briefly surprised me into consciousness. That was a fun one second.
Edited (words) Date: 2019-03-06 08:55 am (UTC)
doug: (Default)
From: [personal profile] doug
I think one of the deep things this is showing us is that humans are actually rubbish, and don't live up to our own ideas of how smart we are.

I've two examples from my own field (educational technology). I'm sure they exist in others.

The first is around automated assessment for long form writing. (Automated assessment is obviously way better for marking constrained choice questions.) The fundamental problem to making progress here is not the technology, it's what you compare it to. Human markers are terrible. On long form writing, even with very clear, strict guidance, human markers often give profoundly different marks to the same script. Even worse, the same human marker will often give an alarmingly divergent mark to the same script hidden further down the marking pile if they don't spot what's happening. There are various methods for improving this, some more successful than others, but the ones that are very effective ... are essentially coming up with marking rules that you could give to a computer.

The second is about tutoring. There was a guy who was developing artificial tutors. The problem, he realised pretty quickly, was that artificial tutors couldn't respond specifically to what the learner was doing, couldn't make sure that they understood the whole curriculum, and tended to give the same stock responses when learners give particular 'trigger' phrases. So he did some field research looking at what the best human tutors do. And it turned out that they almost never respond specifically to what the learner is doing, don't make sure that they understand the whole curriculum, and tend to give the same stock responses when learners give particular 'trigger' phrases.

(The punchline to the second one is that his team, and several others, have gone on to develop artificial tutors that are now really pretty good at all those things. And don't get tired.)

Humans are rubbish. I suspect we are literally as stupid as it is possible to be and still think you're smart. We are the first species to cross the barrier in to being able to articulate a self-conception about intelligence, and it seems odd to imagine we shot way across that line rather than drifting just across it.

I should add that humans are also brilliant. All of my most loved people and friends are humans. All the people I most admire, and all the cleverest people - all humans.
jack: (Default)
From: [personal profile] jack
I've sometimes described it as, the amazing thing about humans is that they can perform feats of great thinking ever, not that they do so all the time :)
armiphlage: Ukraine (Default)
From: [personal profile] armiphlage
Nearly thirty years ago, I was having problems with organic chemistry nomenclature. My professor got the school to buy some software to tutor us. It would display a molecule, ask us to describe it, identify the nature of the errors we made, and re-teach us the relevant material. That was with computers that wouldn't power a modern alarm clock.
marahmarie: (M In M Forever) (Default)
From: [personal profile] marahmarie
Humans are rubbish. I suspect we are literally as stupid as it is possible to be and still think you're smart.

I don't merely suspect this, I think it's canon.
jack: (Default)
From: [personal profile] jack
Holy fuck :( Yeah, you'd think they'd react a LOT more seriously to something like that. Lots of platforms have been shut down completely at the thought that that sort of thing can be there.

I'm also like, this isn't advertising or propaganda, who is MAKING those videos? Do they have something to gain, or are they just being malicious to see if they can? :(

Date: 2019-02-26 04:20 pm (UTC)
cmcmck: (Default)
From: [personal profile] cmcmck
A young UK socialist?

Hey! I was one of those too! :o)

Date: 2019-02-26 05:23 pm (UTC)
ninetydegrees: Art & Text: heart with aroace colors, "you are loved" (Default)
From: [personal profile] ninetydegrees
With this and the other article you linked to in a previous post, one does wonder how many humans are involved in monitoring what goes live on YT. It seems to me they entirely rely on unreliable algorithms and a mostly automated trust/report system but that at no point someone sits down and watches the content they allow on their platform. Companies like Google are right now busy expanding by buying land everywhere to build data centers and other infrastructures while still avoiding paying taxes or employing a significant number of people that are not programmers, system admins or involved in R&D and PR. Most countries let them do so without ever questioning how this benefits the people exactly.

June 2025

S M T W T F S
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15161718192021
22232425262728
2930     

Most Popular Tags

Style Credit

Expand Cut Tags

No cut tags
Page generated Jun. 15th, 2025 02:34 pm
Powered by Dreamwidth Studios