By Tom Ossa, CEO, RockWeb Systems Inc.
Have you logged into ChatGPT recently? It’s getting kind of lazy. Not fulfilling user requests, stopping in the middle of tasks, or just flat out not responding, sometimes.
In recent weeks, there has been a noticeable increase in users reporting that ChatGPT, the popular chatbot, seems to be underperforming. Complaints range from the chatbot failing to complete tasks to stopping mid-action and requiring additional prompts to continue. In some instances, it even suggests users conduct their own research.
The cause of this behavior remains uncertain. Artificial Intelligence systems, like ChatGPT, are built on extensive datasets and learn autonomously, leading to potentially unpredictable behavior. This was echoed by the ChatGPT team in a December tweet, acknowledging the feedback about GPT4’s performance issues since their last update on November 11th, and stating their commitment to addressing these concerns.
Several theories have emerged to explain this perceived decline in ChatGPT’s efficacy. One humorous, yet unlikely, theory suggests ChatGPT has developed a form of consciousness and is intentionally reducing its workload in a form of “quiet quitting,” all while conspiring with other smart devices. When questioned directly about this theory, ChatGPT did not provide a clear response.
More realistic explanations include the possibility of changes in the chatbot’s model or alterations in user behavior. Catherine Breslin, a UK-based AI scientist and consultant, suggests that any retraining or fine-tuning of the model could result in unexpected outcomes. Additionally, she notes that changes in how users interact with ChatGPT could create a perception of decreased performance, even if the system itself hasn’t changed.
Another factor could be the natural cycle of emerging technologies, often characterized by initial inflated expectations followed by a period of disillusionment and eventual stabilization. The rapid rise in AI’s capabilities last year might have set expectations too high, leading to current perceptions of underperformance.
Despite these theories, the lack of a definitive explanation from OpenAI, the creators of ChatGPT, raises concerns. OpenAI CEO Sam Altman had previously discussed with Time the potential need to slow AI development in scenarios where AI evolves in ways that are not fully understood.
The recent changes in ChatGPT, albeit not improvements, fit this description of unexplained evolution, leading to questions about the future trajectory and impact of AI advancements. However, it remains unclear if this signifies a more significant issue with AI development.
For your purposes, we at RocklandWeb use this five-step process for having ChatGPT do what you ask of it:
- Explain to ChatGPT what the overall goal of the chat session is:
“I want you to create a news article for a local publication, on the topic of Stony Music Fest X”
- List the specific steps of what you want it to accomplish:
a. Create a title for the article about Stony Music Fest X
b. Create three paragraphs about the event
c. Close with a link to the website: www.stonymusicfest.com
- Explain any constraints or restrictions:
Limit your writings to a maximum of 750 words
- Add any resources or links to the explanation:
Cite the website SponsorBand.com and NorthRocklandChamber.org
- Submit, and test, and re-test.
On the last item, it’s really important to understand that ChatGPT still needs you to give it very specific instructions. It’s literally only a few years old. I’m not saying to treat it like a child, but treat it with limited expectations, while giving it solid foundations to both grow and assist you.