Chat about ChatGPT

Spent some time playing around ChatGPT a few days ago. Today got some time to write down some interesting things that I found. Really what I want to see is how it can become even more powerful. There are at least a few things that can be done to make it even better.

If ChatGPT can access to the databases in application areas, it can become more powerful.

See examples below:

Q1:

Me: how stock market behaves the two days before lunar new year of 2009?

ChatGPT: I'm sorry, but I am unable to provide a specific answer to your question. As a large language model trained by OpenAI, my knowledge is based solely on the text that I have been trained on, and I do not have access to information from external sources. I do not have the ability to browse the internet or update my knowledge with new information. Additionally, my training only goes up until 2021, so I am not able to provide information about events or developments that have occurred after that date. I apologize for any inconvenience.

Q2:

Me: what date is the two days before lunar new year of 2009

ChatGPT: The lunar new year, also known as the Chinese New Year, is celebrated on a different date each year, as it is based on the lunar calendar. In 2009, the lunar new year would have fallen on January 26th, which was a Thursday. Therefore, the two days before the lunar new year in 2009 would have been Tuesday, January 24th and Wednesday, January 25th. As I mentioned earlier, the date of the lunar new year can vary depending on the specific location and cultural traditions, so it is always best to check with local authorities or sources to confirm the exact date of the holiday.

Q3:

Me: what date is the Christmas holiday of 2009

ChatGPT: In the United States, Christmas Day is typically celebrated on December 25th of each year. Therefore, the Christmas holiday in 2009 would have been celebrated on Friday, December 25th, 2009. Of course, the date on which Christmas is celebrated can vary depending on the location and culture, so it is always best to check with local authorities or sources to confirm the date of any holiday.

Q4:

Me: If the oil price drops 20% within 2 consecutive days, how it is going to affect S&P 500 index

ChatGPT: It is difficult to say exactly how a 20% drop in oil prices within two consecutive days would affect the S&P 500 index, as there are many factors at play. The S&P 500 index is a market-capitalization-weighted index that measures the performance of 500 large-cap companies in the United States, and it is widely considered to be a leading indicator of overall market performance. A 20% drop in oil prices within two consecutive days would likely have a negative impact on the energy sector, which is a component of the S&P 500 index. However, the exact impact on the index as a whole would depend on the specific stocks that make up the index and the broader economic and market conditions. Additionally, a drop in oil prices could potentially have a positive impact on other sectors of the economy, such as transportation and consumer goods, which could offset some of the negative impact on the S&P 500 index.

Comments: I was founding a company, and one important product that we were building there was a NLU chatbot which need be able to answer some similar and more complicated questions like the ones shown above. There exist the following challenges when the model answers this type of questions.

1) The model need be able to specify the specific date according to the description such as holidays.

2) The model need access to the historical database and furthermore it need formulate a right query according the question description to get the right data set.

3) The model needs to construct or specify a right mathematical formula to do calculation on the data set according to the question description. For example, what is the average return of the stock market one week before lunar new year of 2009? The average return is one simple calculation.

4) The model then needs to translate the data result according to 1), 2) and 3) to the appropriate text/charts.

According to the answer to Q1, ChatGPT is not able to address this type of questions directly. From Q2 and Q3, we did see that ChatGPT is doing perfectly OK to specify the date based on the description. From Q4, it is working great to describe the market according to the data and it even explains the casualty between the oil price 20% drop and the S&P index. However, for Q4, we expect to know more. It means that we need it to identify the similar historical events and compare them and summarize the findings. The major reason why it cannot do that is that it cannot access to the historical data.

Integrating ChatGPT with one model similar to image recognition and generation will make it more powerful.

In Q4, instead of asking one question, one chart can be sent to ChatGPT. ChatGPT can call one image recognition to understand what it is, and then answer the question. Or, when ChatGPT need answer one question, it can call one image generation related model to create one chart and use the chart to answer the question besides a text answer.

ChatGPT will become more powerful if it can constantly retrieve the internet and use the new corpus update itself.

Q5:

Me: what is chatgpt?

ChatGPT: I am not familiar with a language model called "chatgpt." As a large language model trained by OpenAI, I am not aware of every model that exists, and I may not be familiar with all of the latest developments in the field of natural language processing. If you could provide some additional information or context, I may be able to provide a more helpful response.

Comments: ChatGPT does not know itself. This is not surprising, however. ChatGPT is a sibling model to?InstructGPT deployed in January 2021. Both of them is using GPT-3 introduced by Open AI earlier in May 2020. At that time, the term ChatGPT is not existing out of 45 TB text data, and the named entity recognition (NER) is not able to work for it.

"OpenAI GPT-3 model has been trained about 45 TB text data from multiple sources which include Wikipedia and books. The multiple datasets used to train the model are shown below:

Common Crawl?corpus contains petabytes of data collected over 8 years of web crawling. The corpus contains raw web page data, metadata extracts and text extracts with light filtering.

WebText2?is the text of web pages from all outbound Reddit links from posts with 3+ upvotes.

Books1 & Books2?are two internet-based books corpora.

Wikipedia?pages?in the English language are also part of the training corpus."

ChatGPT will become more powerful when it is able to access to the third party applications to locate the real-time information.

Q6:

Me: what is the weather today

ChatGPT: I'm sorry, but I'm not able to browse the internet and do not have access to current weather information. As a large language model trained by OpenAI, I do not have the ability to access real-time data or provide information on current events. I am a machine learning model designed to assist with a wide range of tasks, including answering questions and providing information on a wide range of topics, but I do not have the ability to access real-time data like the current weather. Is there something else I can help you with?

Comments: It is evident that ChatGPT will be substantially more useful if it knows how to access to www.weather.com or www.google.com or the traffic related GPS data or the real-time stock market etc. as this will make it have the capabilities to help with the real-time decisions.