eevBLAB 106 - ChatGPT AI Has Changed EVERYTHING

[thm_w]

tldr, not great at quantum physics concepts.

[SiliconWizard]

If you want more "insights" into what MS has in store with ChatGPT:


Nice shit!
Are you sure you still want to touch this crap? I wouldn't even with a stick.

[golden_labels]

I already spot some truly good uses for ChatGPT. Replace the interviewer. After all he does is nothing more than giving Nadella excuses to deliver what boils down to a well-prepared speech.

I feel sad, that the discourse is bouncing between overhyping and fearmongering. Something, that may provide a philosophical shift comparable to theories of evolution, is degraded to such a low topic.

[Fungus]

I already spot some truly good uses for ChatGPT. Replace the interviewer. After all he does is nothing more than giving Nadella excuses to deliver what boils down to a well-prepared speech.

I watched a few minutes of it and the "speech" seemed 100% waffle.

[tszaboo]

BTW how was anyone able to access ChatGPT recently? It always throws a "we are overlaoded, sorry" error whenever I try to access it in the past month. Is this location based?

[PA0PBZ]

I just checked and got the "ChatGPT is at capacity right now" message. I'm sure I've been there in the past month without problems so I guess it's not based on location but on luck.

[madires]

Poor MBAs!

ChatGPT passes MBA exam given by a Wharton professor: https://www.nbcnews.com/tech/tech-news/chatgpt-passes-mba-exam-wharton-professor-rcna67036 (warning: autoplay video)

[ataradov]

MBA exams are supposed to be easy. It is a tool to part people with their money. Only rich people go for it and you can't make exams to hard, or you will fail most of your target audience.

[tszaboo]

MBA exams are supposed to be easy. It is a tool to part people with their money. Only rich people go for it and you can't make exams to hard, or you will fail most of your target audience.

Have you seen the English proficiency tests for entrance?
There is no way I would be passing those tests, and honestly it feels like the design of the test makes it so that foreign students would have no chance of passing.

[ataradov]

Have you seen the English proficiency tests for entrance?

No. Are there credible example test anywhere?

[golden_labels]

Returning to the topic, a friend got a result in which ChatGPT emulated a correct observation regarding its own mistake and provided a good solution. Beautiful! It can mimic the behavior of a human student on an oral exam.

Have you seen the English proficiency tests for entrance?
There is no way I would be passing those tests, and honestly it feels like the design of the test makes it so that foreign students would have no chance of passing.

You are not a native speaker, are you? I guess you would also fail elementary school maths exercise, if I wrote it in Polish and you had no translator at hand, but that has nothing to do with how hard the test is.

[ataradov]

But it is nothing more than mimicking. ChatGPT would accept and agree to your corrections, even if your corrections are completely wrong. This is the method people often use to  make it produce nonsense even more than it would on its own. Fortunately there are solid filters and it is hard to make it go full nazi as Tay did.

[xrunner]

I just checked and got the "ChatGPT is at capacity right now" message. I'm sure I've been there in the past month without problems so I guess it's not based on location but on luck.

It got national news coverage on a major news network in the USA this morning, several times, so perhaps it's becoming even more popular now.

[golden_labels]

But it is nothing more than mimicking. ChatGPT would accept and agree to your corrections, even if your corrections are completely wrong.

There was no correction or any hint. Not even indication, which of the calculations had a mistake. Merely a question, if it can recognize the mistake in “one of its calculations”. Which is what makes it a case different from other outputs expected from a language model.

I just checked and got the "ChatGPT is at capacity right now" message. I'm sure I've been there in the past month without problems so I guess it's not based on location but on luck.

It got national news coverage on a major news network in the USA this morning, several times, so perhaps it's becoming even more popular now.

It already served its purpose too. You don’t think it’s there for you to play forever?

[EEVblog]

BTW how was anyone able to access ChatGPT recently? It always throws a "we are overlaoded, sorry" error whenever I try to access it in the past month. Is this location based?

I get that too. I'm using the a beta version which still has access via my OpenAI login.

[PA0PBZ]

At the moment it's fine again.

[Neutrion]

Why ask a question in Google and then have to search through various resources and weight up their relevance just to get your answer. Let the AI do that.

Sure, right up until the humans start creating web pages specifically to game the AI system.

Of course. Humans already game the Google system.
The "AI" will be an additional layer of opacity, though.

Not just that. Even more people (than who belive that the first  hit in the search engine is the truth) will belive that what the AI said is the ultimate truth because it must bevery smart...

Also to differentiate between humans and bots on the net will become almost impossible at some point.

I already spot some truly good uses for ChatGPT. Replace the interviewer. After all he does is nothing more than giving Nadella excuses to deliver what boils down to a well-prepared speech.

I feel sad, that the discourse is bouncing between overhyping and fearmongering. Something, that may provide a philosophical shift comparable to theories of evolution, is degraded to such a low topic.

I think that the "overhypers" can not really counter the arguments of the "fearmongerers".

I clearly can't even see what is a general aim with a strong AI, apart from "it might will have some good uses". As if it would be some intended God or religion substitution for some. Who is going to doublecheck
the results of the ultimate wisedome?   Or what is 42?
 
The positiv arguments are clearly weak in comparison with the negative ones.

It's like: it could have some good uses  to give a bunch of children a gun because if a terrorist might show up they could have something to defend themself. Yes, but practically it will not do anything good.

For the good of all humanity:   
Actually if we have a look at the last 20 years of internet usage, which was supposed to give freedom of information and equality to everyone the opposite has happened. Even within western societies, differences got bigger, and the general public doesn't seems to be better informed.
Also ruining the planet switched to a higher speed.
But pre-crime and 24 hour surveillance sounds so good...
https://www.theguardian.com/world/2022/jan/10/a-data-black-hole-europol-ordered-to-delete-vast-store-of-personal-data


Now to hope, that if we introduce even more random fast changing factors into the equations will turn things definetively better, is just a belief, without factual or argumental backing.

Sometimes it would be also nice to ask, what is the actualy aim of technical advancement, what do we want to achieve? Is it achievable through technical advancements? Did we get closer to what we wanted in the last....50 years? Or 150?

[tszaboo]

Actually, I was thinking about the possible uses on how this can change our society into something better (I know, right)?
One of the biggest issue IMHO we are facing now is the post-truth society. Fake news and political polarization is a huge issue, and people can live their lives in an echo chamber. The political right and the left both seems to have taken a anti-scientific stance, where they just don't follow scientific evidence. So I asked our robot (which wasn't overloaded tonight, only very slow) if it can do this. It's really reluctant to say that it can detect fake news. And I know the training data ended in 2021 September. But I cannot help but wonder if this can be used to fix this issue. Or if it will make things even worse, since the repeated propaganda pieces will train the model with bad data, and it has no way of distinguishing it from real news.

[Ed.Kloonk]

Actually, I was thinking about the possible uses on how this can change our society into something better (I know, right)?
One of the biggest issue IMHO we are facing now is the post-truth society. Fake news and political polarization is a huge issue, and people can live their lives in an echo chamber. The political right and the left both seems to have taken a anti-scientific stance, where they just don't follow scientific evidence. So I asked our robot (which wasn't overloaded tonight, only very slow) if it can do this. It's really reluctant to say that it can detect fake news. And I know the training data ended in 2021 September. But I cannot help but wonder if this can be used to fix this issue. Or if it will make things even worse, since the repeated propaganda pieces will train the model with bad data, and it has no way of distinguishing it from real news.

Too funny. The same certain peeps in the middle who are poisoning scientific research with 'science', as it were, will end up pumping money and nefarious efforts into this contraption as well. Think Wiki on steroids.

 

[james_s]

I don't think it will either destroy nor revolutionize society. Back in the 60s people were fretting that synthesizers were going to ruin music, they didn't, they just created new genres of music, people still play traditional instruments.

[Ed.Kloonk]

I don't think it will either destroy nor revolutionize society. Back in the 60s people were fretting that synthesizers were going to ruin music, they didn't, they just created new genres of music, people still play traditional instruments.

A better analogy might be Guitar Hero.

[golden_labels]

I clearly can't even see what is a general aim with a strong AI, apart from "it might will have some good uses". As if it would be some intended God or religion substitution for some. Who is going to doublecheck
the results of the ultimate wisedome?   Or what is 42?

Machine learning solutions permit finding answers to otherwise computationally expensive problems. After understanding of how specific ML models work, they may become the next step after current statistics. There is a strong coöperation between machine learning researchers and neuropsychologists, boosting knowledge in both branches. Even the current meme smortnets offer philosophical insights, which may be as crucial and influential as those which led Galileo and Darwin.

I clearly can't even see what is a general aim with a strong AI, apart from "it might will have some good uses". As if it would be some intended God or religion substitution for some. Who is going to doublecheck
the results of the ultimate wisedome?   Or what is 42? (…)

That entire fragment is so all over the place and incoherent, that it’s hard to understand it. But it gives me a strong impression, it’s written as if development was a matter of someone’s choice.

Sometimes it would be also nice to ask, what is the actualy aim of technical advancement, what do we want to achieve? Is it achievable through technical advancements? Did we get closer to what we wanted in the last....50 years? Or 150?

Ponder on that more, taking advantage of not having to pull a plough to marginally offset your hunger and even being aware of such abstract issues.

Actually, I was thinking about the possible uses on how this can change our society into something better (I know, right)?
One of the biggest issue IMHO we are facing now is the post-truth society. Fake news and political polarization is a huge issue, and people can live their lives in an echo chamber. The political right and the left both seems to have taken a anti-scientific stance, where they just don't follow scientific evidence. So I asked our robot (which wasn't overloaded tonight, only very slow) if it can do this. It's really reluctant to say that it can detect fake news. And I know the training data ended in 2021 September. But I cannot help but wonder if this can be used to fix this issue. Or if it will make things even worse, since the repeated propaganda pieces will train the model with bad data, and it has no way of distinguishing it from real news.

Basically as it has been for the past 3000 years, at laest. For an average person with only minor nudges in either direction. With people equally deceiving themselves that “in the past it was better”.

[james_s]

I don't think it will either destroy nor revolutionize society. Back in the 60s people were fretting that synthesizers were going to ruin music, they didn't, they just created new genres of music, people still play traditional instruments.

A better analogy might be Guitar Hero.

Guitar Hero hasn't ruined music either, it's a video game, not a simulator. Nobody goes to Guitar Hero concerts.

[EEVblog]

Guitar Hero hasn't ruined music either, it's a video game, not a simulator. Nobody goes to Guitar Hero concerts.

[bob808]

One of the biggest issue IMHO we are facing now is the post-truth society.

I don't think there has ever been a time when humanity had more access to truth than it does today. The real issue is that it's mixed up with a lot of distortions of it. But we have access to most truth we ever had.
That is why you see the discrepancies of both left/right political spectrums. Because up until the internet started being mainstream they could more easily sell their bullshit.
And anyway, people don't deal that good with truth.