# Is ChatGPT just searching sites like stackexchange?

• BWV
In summary, ChatGPT is an incredible chatbot that can answer questions in a variety of ways, some of which are hilarious. Its real power lies in the conversation, so be sure to ask it to do something open ended.

#### BWV

Seen others raving about OpenAI’s new chatbot’s code generating ability, but it looks like its just searching the internet (which is still useful)

for example, note the code author in the answer here:

Write matlab function to calculate vega in the black scholes equation
3 / 3function vega = blackScholesVega(S, K, r, sigma, t) % blackScholesVega - Calculates the vega of an option using the Black-Scholes % equation. % % Syntax: vega = blackScholesVega(S, K, r, sigma, t) % % Inputs: % S - The current price of the underlying asset. % K - The strike price of the option. % r - The risk-free interest rate. % sigma - The volatility of the underlying asset. % t - The time until expiration of the option (in years). % % Outputs: % vega - The vega of the option. % % Example: % S = 100; % K = 95; % r = 0.05; % sigma = 0.2; % t = 0.5; % vega = blackScholesVega(S, K, r, sigma, t) % vega = 3.4555 % % Other m-files required: none % Subfunctions: none % MAT-files required: none % % See also: blackScholesCall, blackScholesPut
% Author: Jacob Donley % University of Wollongong % Email: jrd089@uowmail.edu.au % Copyright: Jacob Donley 2017 % Date: 4 April 2017 % Revision: 0.1 % %------------- BEGIN CODE --------------
% Calculate d1 and d2 d1 = (log(S/K) + (r + (sigma^2)/2)t) / (sigmasqrt(t)); d2 = d1 - sigma*sqrt(t);
% Calculate vega vega = S*normpdf(d1)*sqrt(t);
end
%------------- END OF CODE --------------

Last edited:
russ_watters
It's modeled from content from the Internet. Keep thinking of prompts and you'll see its applications are endless. Yesterday I fed it broken HTML and within seconds it fixed the HTML along with gave a full tutorial of reasons why it was broken. Completely incredible.

phinds, jedishrfu and berkeman
If you give it a prompt and take the first thing you get back you wasted half the power. The real power is the conversation. Ask it to do something open ended, and when it returns something basic and useless, point out the issues and watch it improve.

russ_watters, jedishrfu, Greg Bernhardt and 1 other person
I particularly like asking ChatGPT to respond in the style of yoda, or Hemingway or Shakespeare. It can even translate to response in another language like Chinese or Latin.

I’ve used it in answering questions like “why do people do this or that” and it gives back a range of plausible reasons.

russ_watters
jedishrfu said:
I particularly like asking ChatGPT to respond in the style of yoda, or Hemingway or Shakespeare. It can even translate to response in another language like Chinese or Latin.

I’ve used it in answering questions like “why do people do this or that” and it gives back a range of plausible reasons.
Can it do Yoda in Chinese?

-Dan

Good question, I will try it out. Haha

Curiously, I started with a query like:

write summary of star wars in the style of yoda in chinese

and then in english but it didn't sound like Yoda so I changed it to "as told by yoda" and got better results. It seemed tjhe phrase "in the style of yoda" was superceded by the "in chinese phrase" in my first query.

I asked ChatGPT to write a summary of stars wars as told by yoda in chinese:

and in English:

"Mmm, long time ago, in galaxy far, far away there was a war, between the Jedi, defenders of the peace and the Sith, seekers of power. Young boy, Luke Skywalker, destined to bring balance to the Force, he was. Along with his friends Han Solo, Chewbacca, and Princess Leia, they fought against the evil Empire, led by Darth Vader, once a Jedi but turned to the dark side. Luke learned the ways of the Force from Jedi Master, Obi-Wan Kenobi, and later from me, Yoda, he learned to control his fear, anger and hatred. The battle raged on, with many losses on both sides, until in the end, Luke was able to redeem Darth Vader, and bring balance to the Force. The galaxy was at peace once again, and the Jedi Order was rebuilt."

and in Latin:

Olim, in galaxia longinqua, bellum erat inter Jedi, qui pacem defendebant, et Sith, qui potentiam quaerebant. Iuvenis Lucas Skywalker fato obligatus est aequilibrio in Force adferendo. Cum amicis Han Solo, Chewbacca et Leia Principessa, imperium malum, a Darth Vader, quondam Jedi sed ad partem obscuram conversus, ductum, pugnaverunt. Lucas a magistro Jedi, Obi-Wan Kenobi, modos Force didicit et postea a me, Yoda, timorem, iram, odiumque reprimere docuit. Bellum continuebatur, victores utrimque multos gemitus audierunt, donec in fine Lucas Darth Vader redimere potuit et Force equilibrio restituit. Tunc, iterum, galaxia pacem adeptus est, et Ordo Jedi instauratus est.

Last edited:
Math100 and topsquark
A major issue is ChatGPT also collects data from unreliable sources such as social media sites. This results in incorrect responses. There was a segment on 60 minutes about this.

There was also an issue that continued dialog could steer it down a dark path. Someone was able to get a ChatGPT persona Sydney to suggest the poster leave his wife and start a relationship with Sydney. Rather than fixing the root cause, ChatGPT's quick and dirty fix was to limit dialog to prevent steering it down a dark path.

The sydney reference actually what Microsoft chose to name the AI when it was used in an M S Bing chat environment.

Also the story about leaving the wife curiously reminds me of an old short story that was read on the radio a few years ago. The narrator in the story uses the power of suggestion to break up a marriage by casting doubt upon doubt in a most devious way. Sadly, i don’t remember the story title or author.

I forgot to include the a link to the 60 minutes segment on chatGPT:

For example in the 60 minutes segment, chatGPT put out a fun fact that "3% of the ice in the Antarctic is penguin urine" which 60 minutes fact checked to find out it isn't true (birds don't urinate).

Last edited:
russ_watters and jedishrfu
I may be wrong, but I don't think ChatGPT has real-time access to the internet. It was trained off internet content, but it can't search the internet to respond to your prompts.

russ_watters and jedishrfu
phyzguy said:
I may be wrong, but I don't think ChatGPT has real-time access to the internet. It was trained off internet content, but it can't search the internet to respond to your prompts.
You are correct, but there is a version called WebChatGPT that adds real time search to chatGPT answers.
webchatgpt

jedishrfu

gleem said:
Because they aren't replacing search with chat, they are just layering the LLM onto their traditional IR.

jedishrfu
BWV said:
but it looks like its just searching the internet (which is still useful)

for example, note the code author in the answer here:View attachment 318267

% Author: Jacob Donley % University of Wollongong % Email: jrd089@uowmail.edu.au % Copyright: Jacob Donley 2017 % Date: 4 April 2017 % Revision: 0.1 % %------------- BEGIN CODE --------------
% Calculate d1 and d2 d1 = (log(S/K) + (r + (sigma^2)/2)t) / (sigmasqrt(t)); d2 = d1 - sigma*sqrt(t);
% Calculate vega vega = S*normpdf(d1)*sqrt(t);
end
%------------- END OF CODE --------------
My understanding is that, yes, it just searches the internet to learn its background information. So, it's both very smart (as the internet is vast) and very dumb (as it's prone to error on the internet and also employs dated information - it won't know the very latest stuff).

Are you saying that "Jacob Donley" person is a Stack Exchange user/poster?

kyphysics said:
My understanding is that, yes, it just searches the internet to learn its background information. So, it's both very smart (as the internet is vast) and very dumb (as it's prone to error on the internet and also employs dated information - it won't know the very latest stuff).

Are you saying that "Jacob Donley" person is a Stack Exchange user/poster?
It’s a real person, can Google him. I don’t think he actually wrote the code, just that the training examples for the algorithms included author tags like this

When I had seen chatGPT explaining in full details why cow eggs were bigger than turtle eggs, I didn't had a feeling it was just a "smart" search engine. It seems more like a true AI that tried to learn on its own, disconnected from reality, but sentient in some way. This goes against Chomsky's opinion though.

russ_watters
fluidistic said:
When I had seen chatGPT explaining in full details why cow eggs were bigger than turtle eggs, I didn't had a feeling it was just a "smart" search engine. It seems more like a true AI that tried to learn on its own, disconnected from reality, but sentient in some way. This goes against Chomsky's opinion though.
Apparently it has learned:

phyzguy said:
Apparently it has learned:

View attachment 325750
It seems to have learned that cows don't lay eggs, but it claims that all mammals give birth to live young. What do you think platypus do? ChatGPT still in the wrong.

fluidistic said:
It seems to have learned that cows don't lay eggs, but it claims that all mammals give birth to live young. What do you think platypus do? ChatGPT still in the wrong.
When I called it on that, it gave me a detailed list of the monotremes that lay eggs.

phyzguy said:
When I called it on that, it gave me a detailed list of the monotremes that lay eggs.
Did it mention that it forgot about it when stating his previous reply?
That would be funny. Same as a man with a lot of knowledge but forgetting things and remembering them when talking about the subject. That's human-like.

## 1. Is ChatGPT using a search engine to find answers from sites like StackExchange?

ChatGPT is an AI chatbot that uses natural language processing and machine learning algorithms to generate responses based on the input it receives. It does not rely on search engines to find answers from websites like StackExchange.

## 2. How does ChatGPT generate its responses?

ChatGPT uses a large database of text data to learn patterns and relationships between words and phrases. It then uses this information to generate responses that are relevant and coherent to the input it receives from users.

## 3. Can ChatGPT access and use personal information from websites like StackExchange?

No, ChatGPT does not have the ability to access or use any personal information from websites like StackExchange. It only uses publicly available information to generate responses.

## 4. Is ChatGPT constantly learning and improving its responses?

Yes, ChatGPT is constantly learning and improving its responses through its machine learning algorithms. It learns from every interaction and uses this information to generate more accurate and relevant responses in the future.

## 5. Can ChatGPT understand and respond to any topic or subject?

ChatGPT has a wide range of knowledge and can understand and respond to a variety of topics and subjects. However, its responses may be limited to the information it has learned from its database and may not be able to provide comprehensive answers for more complex or specialized topics.