Tiernan Ray/ZDNET undertook a comparison between Google’s Gemini chatbot and OpenAI’s ChatGPT-4 with the intention of exploring the concept of ‘stochastic gradient descent’ (SGD) in the realm of modern deep-learning artificial intelligence (AI). In an effort to simplify the complex concept of SGD through metaphor and analogy, Ray posed a question to both programs: Is there a useful analogy for stochastic gradient descent?
The results of the comparison highlighted Google Gemini as the more successful program in providing a satisfactory explanation of SGD via analogy. Gemini offered an analogy involving finding a treasure at the bottom of a valley to represent the narrowing of the gap between desired and actual outcomes. On the other hand, ChatGPT-4 likened the process to a hiker navigating through a foggy mountain descent, struggling to see the path clearly.
Subsequent prompts pushed both programs to extrapolate on their initial analogies. Gemini’s response to a question regarding the placement of the treasure at the bottom of the valley showcased its ability to adapt and enhance the analogy, thus progressing the conversation. In contrast, ChatGPT-4 faltered in its response, failing to improve upon its initial explanation of the analogy.
Overall, Ray’s experiment suggested that Gemini exhibited a higher level of responsiveness and adaptability compared to ChatGPT-4 when presented with challenging prompts. Given the significance of SGD in deep learning AI, the conclusion drawn was that Gemini emerged as the more effective chatbot in this particular test.
