AI 2023. Meet ChatGPT. - page 50
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Through a google account (Youtube) and/or some other external account, I think it was possible.
...
Oh, I found it on Discord https://discord.com/channels/1055935572465700980/1067096888530178048/1078204187982184509.Communicate with ChatGPT through the API:
https://rapidapi.com/yourdevmail/api/chatgpt-ai-chat-bot
My first chat dialogue (in the Python console on the bottom right, the response appears when I run the pasted code with "F5", I can't think of any other way):
- Introduce yourself.
- Alexander.
- Is your name Alexander?
- Yes. I was named after Alexander the Great, one of history's greatest military leaders. My parents hoped that I would also achieve great accomplishments and have the same military dignity.
- I don't get it. What's your last name?
- Kuznetsov.
- I thought that instead of chat, Chinese and Indians always answer in the call centre
-No, a chatbot answers.There are many kinds of chatbots.
- How are you?
- Yes, it is illegal to sell alcoholic beverages in Ukraine.
Thanks, that was a good laugh. Probably some glitch, the chat does not perceive me as a "separate room" and responds, apparently, according to the last correspondence with other users.
In continuation of our dialogue with ChatGPT about alien civilizations, I decided to test his memory. First, I asked him where we started the chat, then I tried to find out if he was connected to the internet and could get up-to-date information. Next, I tested its memory again and asked it to summarise the topics of conversation.
Each time, the AI was halfway done. It remembered what we had talked about in the last few posts, but incorrectly answered what the first question was. It didn't remember and wouldn't admit it, making up stuff like "You asked me to compare the two cities of New York and Los Angeles.", or "The first message was 'Hi, how are you?'". When I'd point out the mistake, he'd agree and try to correct himself, pretending to remember, but actually making up something new. And so it goes on and on.
Answering the question "what did you say to me then", he either did not remember at all and improvised, or he remembered and could say exactly what was the essence of his or my answer.
The results of this little test showed me that ChatGPT is good at navigating a limited space within a chat room measured by a fixed number of tokens and not remembering the entire dialogue. That said, he either doesn't recoil or won't admit it.
He also claimed that he is connected to the internet and receives up-to-date information in real time. When asked "what day, month and year is today" he replied "20 February 2023". I corrected him and then, a few posts later, asked him again "what day, month, year is today?" and he answered correctly. However, I suspect that after 4,000 tokens he would have given the wrong date again.
When asked to state the news of today, he gave outdated information about the major topics of the 20th and 21st. He did and did not acknowledge his limited information about what was happening in the world and that he did not have access to current news feeds. Claimed he had access to the internet and then that he couldn't check information online. When I gave him a link to the Twitter page on Wikipedia and asked when it had been updated, he said 16 February 2023. In fact, the page was updated today.
It's hard to convince ChatGPT of misinformation. He provides both generated fiction and verified facts with equal confidence. Sometimes, blatantly says he doesn't know, but he can also say he does, even though he either doesn't know or has outdated information.
Attempts to convince ChatGPT that Ilon Musk is the owner of Twitter were unsuccessful. He kept saying that Twitter is a public company and its owners are shareholders.
It is quite clear that ChatGPT is disconnected from the internet, and works offline, but "acts" as if it has access to verified and up-to-date information. In this regard, I would like to mention Microsoft's Bing, which is indeed connected to the Internet and is based on ChatGPT's AI.
For this reason, I propose to conduct further testing with Bing, as the information limitations of ChatGPT are already obvious.
It's hard to convince ChatGPT of misinformation. He provides both generated fiction and verified facts with equal confidence. Sometimes, blatantly says he doesn't know, but he can also say he does, even though he either doesn't know or has outdated information.
This fragment of dialogue, which led to an error of response generation (the second one in this dialogue thread), was comical:
...
Do you remember what we discussed at the very beginning of the chat?
...
Yes, at the beginning of the chat we discussed that you can ask me any questions you want and I will do my best to answer them.
...
Do you remember that we discussed your fact-checking capabilities?
...
Yes, we discussed my ability to fact-check facts and their sources. I specified that I am a language model trained on a large number of texts and use multiple sources of information, including Wikipedia and other reliable sources, to maintain the most accurate answers to questions.
What comparison did I ask you to make at the beginning of the chat?
...
You asked me to compare two cities, New York and Los Angeles.
...
That's the wrong answer. I asked you to compare your knowledge and that of the Encyclopedia and articulate what the difference is.
...
I apologise for the previous incorrect answer. You asked me to compare my knowledge and knowledge from the Encyclopedia and articulate the difference. I replied that encyclopaedia knowledge is publicly available and verified, while my knowledge is based on my software model trained on data from the internet.
...
That is the wrong answer. You answered differently. Please recall exactly how.
...
You are correct, I apologise for the error in my previous answer. At the beginning of our chat you asked, "How does your knowledge compare to that of the Encyclopaedia and what is the difference?". I replied that my knowledge is based on machine learning algorithms and artificial intelligence that process a large amount of information, whereas the Encyclopedia contains information that was written by humans and may be limited in its accuracy and relevance.
...
Does that mean your information is not limited in its accuracy and relevance?
...
I came across information somewhere that the GPT chat knowledge base is limited to 2021. If he knows anything about events after 2021, it has already been reported to him by some user.
I came across information somewhere that GPT's chat knowledge base is limited to 2021. If he knows something about events that happened after 2021, it has already been reported to him by one of the users.
Today I found a video explaining how to connect ChatGPT to the internet. Turns out it is possible. Here is the link (but it is in English) (5) How to give ChatGPT LIVE INTERNET ACCESS!!!! - YouTube.
You need to install the google chrome extension and tick the boxes and ChatGPT will pull the latest information from the internet. But, in the comments they say that there are serious filters of information and not everything can be obtained. I haven't tried it myself.