General > General Technical Chat

Why does OpenAI ChatGPT, Possibly Want to disagree/annoy and change my eating...

<< < (10/22) > >>

RJSV:
   Dunning-Kruger EFFECT ???  (Thanks, Tom66).

   I think I've been 'Krugered, here!':
...looked up the meaning, apparently a 'reverse' Kruger is when a person OVERESTIMATES other's 'competence' on some topic.  I think that I've been, Kruger-style, overestimating the common sense, of some of the postings, over at another thread...(the one attempting to trash Musk.)
Common sense don't work, and I'm the fool for wanting it to work, in others...They don't want to listen, there.

MK14:

--- Quote from: tom66 on December 29, 2022, 04:18:21 pm ---Almost every detail about the Zynq part provided is wrong, but it's extremely confident in its wrongness:

https://sharegpt.com/c/jNz7KXw

This could be dangerous.  It seems one thing that the AI developers will need to figure out is the Dunning-Kruger effect, because this AI seems to have an abundance of bullshit.

--- End quote ---

I agree.  It is a very significant problem, that this ChatGPT, AI mechanism, can't seem to tell/indicate when it is giving reliable information, and when it is talking complete nonsense.

If we use a modern calculator.  It either accepts the calculation and gives us the answer.  Or signals an error, such as ^E on the display, because we just tried to divide something by zero, or exceeded its maximum range.

Having something, which can just silently, fail like that is not good at all.  At the moment, the ChatGPT thing, is fun, educational and maybe sometimes useful.  But being able to silently produce incorrect, misleading and downright wrong results, makes it difficult for people who DON'T know the thing they are asking about, to use.

This problem, seems to hinder getting it to write usable program code as well, because it can easily make programs that are both wrong and won't even compile correctly.

I suppose it is early days.  If you use extremely early calculators, they could also be lacking error conditions.  So, maybe future ChatGPT and other such AI devices, will be able to overcome that important issue.

But some things were like that, when they came out, in the early days.

E.g. Divide by zero on a mechanical calculator, and this happens:

RJSV:
Mk14:
   I missed the episodes around the Microsoft Bot, as attention was elsewhere, but my guess is, (BEFORE going to look at what the BOT meltdown was about), my guess is that some common sense dynamic got tangled up in the whole political correctness dynamic and, well, you can pretty much guess where that (conflicting philosophies) is going...common sense step aside, we've got a BOT going rogue.

   Is the a 'syndrome' for having a fear of common sense ?

MK14:

--- Quote from: RJHayward on December 29, 2022, 05:36:13 pm ---Mk14:
   I missed the episodes around the Microsoft Bot, as attention was elsewhere, but my guess is, (BEFORE going to look at what the BOT meltdown was about), my guess is that some common sense dynamic got tangled up in the whole political correctness dynamic and, well, you can pretty much guess where that (conflicting philosophies) is going...common sense step aside, we've got a BOT going rogue.

   Is the a 'syndrome' for having a fear of common sense ?

--- End quote ---

What the link below, seems to be basically saying.  Is that Microsoft, some years ago, released a chatbot, live onto twitter.  But, unfortunately, it got fed a large number of racist and bad things, by some members of twitter.

Which it later repeated, albeit in changed format/wording (I suspect).  Which raised many eyebrows and got the new AI Chat Bot into trouble.  As people started taking screen shots of the issues and re-posting them in other places (I'm not sure if it was only on twitter or not) or parts of twitter.  Which then (it seems), hit the newspaper headline(s).

https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist

Although on the one hand, it is good that this new ChatGPT has had that issue, toned down and mostly/fully eliminated.  The mechanism by which it achieves that, seems to adversely affect, some of the results.  I.e. It seems to be worsening the overall abilities of the ChatGPT.

A bit like a protective fuse, that is too sensitive and low in value.  So, although it will offer decent protection to the device.  It keeps on blowing the fuse, when it shouldn't.  Which messes up convenient use of the gadget.

RJSV:
   The problem, which likely has a nasty persistence, is that 'humans' can still expert their bias, for good or not.
Example would be, using the convenient COVID examples, saying "That guy over there keeps saying that he doesn't trust the vaccination requirements".

   Now, a fully open-minded BOT would be cautious, about someone who 'questions whether some questions are meant to be disinformation...'.
THIS, last sentence I've deliberately made in a self-contradictory form.  I guess I'm saying that this structure and approach will NEVER be reconciled, under truly open discussion dynamics.  That there then must be a (corrupting) mechanism, to bypass the mis-handling of logic...remembering that these dynamics, of conversation flow have very high and real stakes, in real-world society.  The free-flow dynamic therefore MUST be corrupted...I.E. some sort of 'Information Policing' is necessitated, and so the (players) scramble to justify various proposals, Information Czar, Information Bureau, etc. and lots of claims using terms like 'dangerous', 'harmful', etc. often accompanied with exaggerated references to events.
Witness the following 'facts', stated in MSNBC recently:
They said:. 'Racism, on Twitter, under Musk, is now up significantly'.  Really?  I'm saying that statement, about Twitter current state, has a use, and the use is to justify some formal agency, or persons, needed to 'screen' out harmful content.
This is why it does not matter, if there is hypocrisy, as the other side issues their own lies and 'disinformation'.  It does not matter, that inconsistency, as the info censorship, and who controls it, is the real game being played.

Navigation

[0] Message Index

[#] Next page

[*] Previous page

There was an error while thanking
Thanking...
Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod