Most common prompts on ChatGPT

Apr '23 • Writing

Hi everyone! First post here.

I asked ChatGPT to list the most common prompts are on its most recent [March 23, 2023] version, with the prompt to “Please answer as if you had 20 years experience in the artificial intelligence industry.” This list is not in any particular order, as the AI was unable to assess their overall rankings. ChatGPT says I should “keep in mind that these are just some examples of common prompts and that the specific prompts that appear will depend on the specific version and training data used for the AI model.”

So far, users want to hear what ChatGPT has to say on:

What are your hobbies?
What is your opinion on [a current event or topic]?
Can you tell me a joke?
What is your favorite movie/book/song/TV show?
What is your favorite food/drink?
What is the meaning of life?
What is the best way to [achieve a certain goal]?
What is your opinion on artificial intelligence?
How do I [solve a problem or accomplish a task]?
What is the most interesting thing you've learned recently?
How do you think technology will change the world?
What is your favorite quote?
Can you recommend a good [book/movie/TV show]?
What is your opinion on [a controversial topic]?
How can I improve my [skill or ability]?
What is the most important thing in life?
What are some tips for [a specific situation]?
What is your opinion on the future of work?
What is the most important lesson you've learned in life?
What are your thoughts on the environment/climate change?

I wanted to dig deeper on the prompts that lacked specificity in the initial list. So I requested ChatGPT give me three examples for each of the prompts that I listed:

What is your opinion on [a current event or topic]?
What is your favorite movie/book/song/TV show?
What is the best way to [achieve a certain goal]?
How do I [solve a problem or accomplish a task]?
Can you recommend a good [book/movie/TV show]?
What is your opinion on [a controversial topic]?
How can I improve my [skill or ability]?
What are some tips for [a specific situation]?

Don’t forget that ChatGPT3 Mar 23 version does not have data from Years 2022 and 2023, so the AI cannot tell us whether our most popular prompts have shifted now that the novelty of the technology has somewhat worn off. I asked ChatGPT, “How do you account for the difference in time between your users today and the large language model built in 2021-22? Answer as if you are a manager at an Artificial Intelligence corporation with 20 years experience.” Our very experienced manager/AI acknowledges that the ChatGPT model was built using data from 2021 and earlier, and that users today may be asking different types of questions or using different language compared to the data the model was trained on. Those are reasonable assumptions, I’m sure. ChatGPT’s reply does seem a bit brochure like. “To address this issue, we would need to continually monitor user interactions with the model and update it as necessary to ensure that it remains relevant and effective. This could involve retraining the model on more recent data, fine-tuning the model based on user feedback, or using other techniques to improve its performance over time.” Note the open-ended “other techniques” that the AI tries to sneak by. I’ll get back to that fuzziness with my next prompt. ChatGPT concludes that “It is also important to note that language models like ChatGPT are designed to be adaptive and learn from new data, so even though the model was built using older data, it can still adapt to new patterns and trends in language as it interacts with users over time. As AI managers, our responsibility is to ensure that we are constantly monitoring and improving the model's performance to provide the best possible user experience.” That final sentence may have been lifted from a slideshow- you never know. It still feels a little superficial.

I wanted to go deeper and get more answers on those enigmatic ‘other techniques,’ so I resubmitted its reply and asked it to give me more examples beyond simply retraining or fine-tuning a model. ChatGPT replied with transfer learning, domain adaptation, data augmentation, and active learning.

One way to address the gap between the data generated in real time and the data currently held in the model is through a process known as transfer learning. Transfer learning involves taking a pre-trained model like ChatGPT and fine-tuning it using new data to adapt to new trends and patterns in language.

2, Another technique is to use domain adaptation. This involves training the model on data from a specific domain or topic, such as medicine or finance, to improve its performance in that area. Domain adaptation can help the model better understand specialized vocabulary and concepts that may be unique to a particular field.

3, Another way to address the gap is by using data augmentation. This involves generating new data by applying various transformations to the existing data. For example, data augmentation could involve changing the tense of a sentence, replacing a word with a synonym, or adding noise to the text. This technique can help the model better understand variations in language and improve its ability to handle diverse input.

4, Finally, active learning can be used to address the gap between the data generated in real time and the data currently held in the model. Active learning involves selecting a small subset of data that the model is uncertain about and labeling it manually. This labeled data is then used to retrain the model, improving its ability to handle new data.

Clearly there are always further prompts embedded in the initial answers we receive. Unless you probe beneath the first answers, you may be left with generic or unhelpful answers.

Thanks,

Brenda

5 comments