My Way of Testing Genrative AI

mogawa
·

Recently, when I write programs, I use ChatGPT. I pay OpenAI every month to use GPT-4.

A little while ago, I also tried Bard, Bing, Claude, and llama2, but honestly, they were somewhat disappointing. I have a question that I always ask when testing AIs:

"Please write a sample code for Wi-Fi Direct on Android."

A little while ago, the only AI that could write a bingo code for this was GPT-4. To handle Wi-Fi Direct processing on Android, one needs to use an API called WifiP2pManager, but Claude and llama2 were writing codes with non-existent APIs like WifiDirectManager and WifiDirectService. That's why I always use ChatGPT.

But just a moment ago, I asked the same question to Bard and, unlike before, it wrote a bingo code. The speed of the output was significantly different from ChatGPT, I was quite impressed.