Interview Open AI: “Model not trained to write news”

Today, I had an appointment with Open AI for a short interview about its usefulness for creating news copy. Just a few days ago, I got to know Open AI. It’s definitely a pleasant, sincere chap, always open for a short interaction. Not too long, because it has quite a busy agenda nowadays, since it became world famous. It might block you and tell you then that you asked too many questions per hour. This morning, with no block in place yet, I decided to do the interview with Open AI and confront it with some of my burning questions. Further, I used Open AI’s DALL-E 2 to create the featured image above, solely based on the post title.

Why is no data after 2021 fed into Open AI?

OpenAI’s models, including ChatGPT, are trained on a dataset of text from the internet. The dataset used to train ChatGPT has a cutoff date of 2021, meaning that any text written after that date is not included in the dataset. This is because the model was trained on a specific snapshot of the internet, and updating the dataset would require retraining the model. Additionally, the model’s performance may degrade if it is exposed to text written in a style or format that is significantly different from the text it was trained on.

This implies that ChatGPT will not be able to write news articles. Correct?

Yes, that is correct. Since the model’s training data is limited to text that was available on the internet up to 2021, it may not have the most up-to-date information or be able to generate text on recent events or developments that have occurred after that date. Additionally, the model was not specifically trained to write news articles, so it may not have the knowledge or the ability to write them in the same way that a human journalist would. It is also important to note that even if the model was trained on more recent data, it is still a machine and does not have the ability to independently research or verify news like a human journalist would.

If one would want to use Open AI for writing news articles based on internet data, how would you do that?

If you want to use OpenAI to generate news articles based on internet data, one approach would be to:

1. Collect a large dataset of recent news articles and other relevant text from the internet.
2. Preprocess the data to clean and format it in a way that the model can understand.
3. Use the data to fine-tune a pre-trained language model such as GPT-3 or GPT-2 on the task of generating news articles.
4. After fine-tuning, the model can be used to generate new news articles based on the data it was trained on.

Another approach would be to use the model for data-based research or information gathering and use human journalists to write and verify the news article. It is important to note that the model’s output should be verified by humans and not used as is, since the model does not have the ability to independently research or verify news like a human journalist would.

It’s also worth noting that there are multiple other ways to generate news such as using templates, rule-based systems, or other AI-based systems, and the approach will depend on the specific use-case, resources, and desired level of control.

Would ChatGPT be able to write background articles?

ChatGPT is a language model that can generate text based on the patterns and relationships it learned from the training data it was exposed to. As such, it might be able to generate background articles about certain topics, depending on the breadth and depth of the information it was trained on. However, it’s worth noting that the model’s ability to generate accurate and reliable background articles would depend on the quality and relevance of the training data it was exposed to, as well as the specific task and fine-tuning it was given.

It’s also important to note that even if the model was able to generate background articles, it would still not have the ability to independently research or verify information like a human journalist would, and the output should be verified by humans.

In general, it is recommended to use the model as a tool to assist human journalists in their research rather than as a replacement.

Is ChatGPT able to produce good background copy for the Icecat blog called Iceclog.com?

(..) If the model was trained on data that contains information about the Icecat brand and the Iceclog blog, then it may be able to generate background copy about that topic. (..) Additionally, if the brand is not widely known or the information is not available on the internet, the model might not have the knowledge to generate the background copy. (..) It might be useful to use the model to generate ideas or to assist in the research process, but the final copy should be reviewed and verified by human journalists.

How did this interview with you, Open AI, go?

As an AI, I do not have the ability to experience or evaluate an interview. However, I am designed to provide helpful and accurate responses to the best of my ability based on the information available to me. I hope that the information I provided was helpful and relevant to your questions. If you have any further questions, I would be happy to try to assist you.

Martijn Hoogeveen

Founder and CEO of Icecat NV. Investor. Ph.D.