The biggest problem with most AI

Lounge
1

To me, appears to be the input. They are relying on the internet for input. A very unreliable source of information to say the least. A human collection of knowledge which is very useful if you have the ability to discern what is reliable information and what is not. The ability to check multiple sources and determine what the general consensus is also usually right. Though not always.

I think AI might be more useful if it were to use only the most reliable sources of info to pull from. Maybe it would not be able to answer such a broad scope of questions all the time, but it might be more reliable.

Anywho, those are my thoughts atm.

6 Likes
2

Thoughts, anyone?

Am I wrong or am I right?

Whats your opinion?

Alexa is particularly bad at this. Amazon often uses contributers answers and spits them out, which are often either wrong, incomplete or half right.

I’ll give you an example. A few weeks ago I was thinking the statue of liberty was made of copper, but was not sure.

Alexa answered that it was made of steel.

I rephrased the question and she/it responded that it was made of copper.

The truth is that the statue of liberty is a copper statue with a metal framework.

1 Like
3

Wikipedia was once considered an unreliable source too.

It wont be long till robots takeover and start to harvest our hearts :robot: :heart: :purple_heart: becuz its only gonna get smarter and more accurate with more feedback/data.

2 Likes
4

ok, no one wants to bite on this topic.

But I don’t remember the T-800’s or T1000’s in the terminator series mistaking foam for metal and trying to take out Sarah Conner with a nerf gun.

So maybe we need to ditch some of the sources of information. All Im saying.

1 Like
5

I don’t think input is an issue, I think verification of the info is an issue. AI can’t think for itself. It can proofread but for example for advanced physics it has no sense of intuition !

2 Likes
6

Maybe you’re right. I am not an expert on AI. In my youth I did some programming, though.

And the saying “Garbage in, garbage out” seems like it might apply.

But like I said, not an expert on AI.

1 Like
7

I only noticed it during the physics degree I was doing. I’d put info in, with as much context as possible but it’d still come back with nothing but confusion - as you’d expect

And yeah, I remember you mentioning you did some programming :fire:

1 Like
8

Well right now the problem is they seem improve with scale, so the more data the better. I think they were saying they’re actually running out of internet to train the things on. I don’t know if they’ve done it yet but they’re talking about creating and using synthetic data, which in theory could be of any quality. As the algorithms improve though it’ll be less and less about scale. It’ll be interesting to see how Qstar works.

1 Like
9

I guess part of it may be whether you want AI to spit out accurate information or are looking at the more futuristic aspects of real learning.

I can’t see how adding false data can improve its accuracy in giving factual information to humans, at least in the short term.

I can, I guess, see how it might help a bit in terms of long term growth.

1 Like
10

When you say false data are you referring to synthetic data?

1 Like
11

no. I dont know that much about synthetic data, besides the definitiion I looked up on the internet. I mean , lots of stuff on the internet is not accurate or is opinion.

2 Likes
12

Oh ok I see what you’re saying. In theory synthetic data could improve the accuracy of the models responses. If we can imitate high quality data and produce a lot of it then that might make the models more reliable. But like you said I’d rather they focus on actually making the models better at learning then just making them better at what they do now.

1 Like
13

Have you tried Claude 3.5 sonnet?

1 Like
14

I think AI can be a neat tool to summarize info on the web. Google is typically good at that.

But like you said, @Bowens , the web isn’t always a reliable source of info.

I think we’re a long ways away before AI becomes sentient in any way.

2 Likes
15

I have not.
15AI’s

16

I think it’s the most capable model right now if you want to see if it’s any better then what you’re using now. I read a lot about ai but actually only using bing lol

1 Like
17

I will take a look sometime soon. Thanks!

1 Like
18

Depending on the AI you’re using, you could always ask to use certain sources that you trust, using internet is not really a problem in my opinion because the Internet does have reliable sources, you just got to know where to look, it’s perfectly acceptable in school to use the internet you just got to make sure you’re using reliable source.

1 Like
19

you’re right, but that was exactly my point, you and I as humans can be pretty adept at determining what info is reliable by comparing sources and different info. I’m not so sure AI is capable of doing that proficiently at this point.

1 Like
20

I mean couldn’t you program the AI to prioritize sources from people who are experts in the field, I mean, there’s literally websites that are dedicated for only people who are experts to publicate on a particular field.

1 Like