ChatGPT: Next Level Dialogue or Uncanny Valley?
ChatGPT is fine-tuned from GPT-3.5, a language model trained to produce text. ChatGPT was optimized for dialogue by using Reinforcement Learning with Human Feedback (RLHF). RLHF is a method that uses human demonstrations to guide the model toward the expected and desired behavior. The AI model feels incredibly realistic and natural because the models were trained on vast amounts of data that was written by humans. The data sets included conversations so that the AI learned responses that sound conversational, human like.
I took ChatGPT for a spin and asked a few business-related questions. I started with a general marketing question and asked ChatGPT to come up with a tagline for "Data Conversion Laboratory."
Surprisingly, these suggestions are very close to taglines, headlines, and marketing campaigns that DCL has used over the past 41 years. Not exact but very close. Because these tagline suggestions were so close I looked into the FAQ section on the website. I thought that a web search of our company name could lead an algorithm to this kind of response. However, ChatGPT is not connected to the Internet so the results were not directed by previous DCL marketing:
"ChatGPT is not connected to the internet, and it can occasionally produce incorrect answers. It has limited knowledge of world and events after 2021 and may also occasionally produce harmful instructions or biased content."
I continued my conversation and asked a question we often answer in our webinars but I put a little spin on my phrasing and asked it to explain the concept of XML to me like I'm a 5-year-old:
The conversational tone comes through loud and clear in this response. I asked Google the exact same thing and would have to click through a number of results with none of them responding in a succinct way that is quickly understood with a meaningful example.
To continue this conversation, I asked what are the benefits of XML:
Right on! Direct and easy to understand.
OK, let's dig deeper and ask the big question:
Nailed it! Though ChatGPT could have simply answered "42" and left it at that.
OpenAI has made it very clear we still need to be careful with the answers provided but that's no different than the wider web. I'll continue my experimentation with the various AI applications and continue to refine what I think about practical uses of AI.