[kwlug-disc] ChatGTP + Bing: we're entering a new paradigm and it's mind-blowing

[Ronald Barnes]

Khalid Baheyeldin wrote on 15/02/2023 11.53:

> The one thing I see that is a huge leap is the user interaction with the 
> AI: that is where the innovation has progressed.

Agreed, it's remarkable.



> Those who are clamouring about AI's capabilities are non-specialists 
> (mainly tech and non-tech journalists, Youtubers, ...etc.)

No, that's just wrong.

Google, Microsoft, etc. are hardly non-specialists and they're all over 
this. Google had an emergency meeting, bringing in the retired founders, 
and put together a disastrous, hasty, ill-fated product demo in response.

They're suitably scared right now.


> Here are two videos by two professors, one in physics and one in 
> astophysics.
> 
> https://youtu.be/GBtfwa-Fexc <https://youtu.be/GBtfwa-Fexc>

So, the physics prof said it got the first questions reasoning correct 
but the answer wrong. Wrong in a manner that he often sees with 
students, and is guilty of himself.  I think he said a 2/3 score on that.

Then an entirely wrong answer.

Then, wrong reasoning but correct answer.


Still seems like leaps and bounds above anything that came before (it 
showed its work, for example), and it's still only 2½ months into a 
public beta.




> https://youtu.be/K0cmmKPklp4 <https://youtu.be/K0cmmKPklp4>

First 2 were exactly correct.

Third correctly references Kepler's third law, but gives the wrong answer.

It got 9, 10, and 11 correct, but we don't see them.

Question 12 is right, so one wrong so far.

Question 13 it got wrong (same answer I gave, oops).

14 is right. 15 is wrong, and I got it wrong too (I guessed rapid rotation.)

16 wrong. It expects multiple choice questions to have only one correct 
answer?

17 gives 3 correct answers, nice.

45½ / 60 so far is "not bad" but "below initial expectations". Way 
better than me though.

Next question is correct, way over my head.

Next one is a units problem, and wrong. I have no idea at this point. 
Interestingly, it said, "This is a very small radius." Yeah, way too 
small, apparently.


66½ / 90 or 73.9%.  Just below the average Columbia non-science major 
student, on an open-book, calculators allowed test.



The fact that the software can "understand" the questions asked 
impresses me.

Feels like were at the "Google invented page rank" stage, when Google 
Earth and Google Street View are still unfathomable.



Note, I'm still enjoying the awe of the achievements of ChatGPT.

I won't be surprised when, like the "Information Superhighway" became a 
misinformation / disinformation sewer, the GPTs of the world are used to 
sell more advertisements, and worse.


rb