What are people asking to ChatGPT? An insight into 10 thousand chat messages between humans and AI assistants.
What’s hot these days?
“Spam calls” on WhatsApp? Crisis in Pakistan? Election in Karnataka?
Not sure, but in technology, it’s certainly- “Tech layoffs”, “Funding winter”, “Valuation slash of Unicorns” and ongoing debate on whether “AI will take over thousands of human Jobs”.
Let’s talk about AI and GPT. On May 10, at Google I/O, Google made Bard, the company’s generative AI chatbot, more widely available, including in India. While OpenAI’s ChatGPT is already disrupting the industry. Microsoft announced Copilot, and tools like Auto-GPT are taking over the Internet. But what are people expecting from these tools? What are users asking to Chat GPT?
Recently I released a GPT-3.5-turbo-powered AI bot on Android Playstore called “Pocket AI”. In this app, users use their own Open AI key to talk to an AI assistant. In this blog, I am sharing insight into user's queries to bot.
In case you are wondering about data privacy, this app only collects user queries and no personal data. This disclaimer is already mentioned in the help section so users are fully aware of this. Also, this is an open-source app and up-to-date code is available on GitHub.
Pocket AI app Metrics
This app is available on Playstore and it has 500+ downloads. The below data shows the distribution of users across countries-
Chat messages Insight
Now that introduction and disclaimer are over, let me directly share insight into users’ queries. I analyzed roughly ~13K messages and feed them to ChatGPT for classification. The messages which were incomplete or inappropriate were filtered out. Here is the outcome-
Classification Categories
Below is the explanation of the categories-
1. Task-oriented conversation
Conversations related to a specific task or goal. Examples-
“Asking for help with a homework problem”, “Asking how to do something, “Asking for advice on a specific topic” etc
2. Informational conversation
Conversations where the main purpose is to exchange information. Examples-
“Learning about a new product”, “Researching a topic”, “Getting expert advice”, “General knowledge and trivia”, “Science and technology” etc
3. Social conversation
Conversations that are primarily for socializing. Examples-
“Chatting about hobbies”, “Talking about current events”, “Making jokes”, “Playing games” etc
4. Personal advice and self-improvement
Conversations related to personal growth or improvement. Examples-
“How to be more productive”, “How to manage stress”, “How to set and achieve goals”, “Health and wellness” etc
5. Philosophy and morality
Conversations related to philosophical or moral issues. Examples-
“Ethics”, “Religion”, “Meaning of life”, “Social and cultural issues” etc
6. Business and Finance
Conversations related to business or finance. Examples-
“How to start a business”, “How to invest your money”, “How to save for retirement” etc
7. Language and communication
Conversations related to language learning or communication. Examples-
“How to improve your writing skills”, “How to improve your public speaking skills”, “How to learn a new language” etc
8. News and current events
Conversations related to current events or news stories. Examples-
“Politics”, “Economics”, “Social issues” etc
Conclusion
As evident from the Pie chart, the categories with the most messages are“task-oriented” and “informational”. These are then followed by “social”, “personal advice and self-improvement” and “philosophy and morality”.
If you want to take a look at the data by yourself, here is the link: https://gist.github.com/varunon9/0ca4868895c5e2baacf793aeb7304c5c
