Yesterday I broke ChatGPT a couple of times. In one occasion I asked it to generate a text matching a regular expression and it did a pretty good job. Then I asked it to explain the regex in plain English. it was correct 80%. This means it was wrong in one statement out of the total five statements. For me 80% correctness is good enough.
in the second instance, I asked for some code and it was unable to provide proper use of the external API that I wanted to be used in the code. When I asked to correct it ChatGPT was unable to properly point to the right API. However, later when I ask it to write the Python script without the API it did a pretty good job. Overall, that was about 60% correct and saved me about 50% of my time. that too is comparatively good productivity.
As of late I have seen lots of criticism on the correctness of ChatGPT. But most of these failure example cases are hypothetical than real. I use ChatGPT regularly for most of my work and I think that makes me twice productive. And that is good enough for me for the time being.
I'm in the process of building some advanced tools using the Open AI API to utilize the GPT model. And I'm hopeful that these tools that I'm building will make my life far more productive.
The hype ground AI keeps growing day by day. Some say that you don't need to worry about learning or understanding AI even though the trend is super hyped. In a true way. However the more you understand the tool the better utilization that you can yield out of those. I think the new usage patterns require novel thinking.
We are trying to figure out how to get these tools to be utilized in the project based learning system that we are using Hindi Avinya Academy. Not only those who want to do The IT curriculum but everyone in general no matter what vocational program they want to use we will give them exposure to these AI tools.
In my personal opinion ChatGPT he's very good at generating textual content based on quality prompts that you provide. Being able to articulate quality prompts is a skills itself. I also know that some people are skeptical if the chat model would be able to replace the search model. This is based on all the hype around Microsoft Bing planning to use ChatGPT for their search engine. For many years I have since the lack of intelligence on the part of search engines to find what we really need. Maybe it is because the search engine revenue models are heavily dependent on the advertising model, for example, Google. Therefore it is good that Google is being challenged by Bing for a market share in the search space with the ChatGPT model. Only time will tell how successful all these going to be. However, I'm hopeful that we will find information easier online with better relevance to what we are looking for.
Today I also tried to draw a complicated picture with Dall-E. Here is the prompt that I used.
“Make a poster with a lotus flower in a tranquil pond at the bottom, surrounded by lily pads. In the center-right, show a boy and a girl looking up towards the top left corner where a modern city is placed. On the left side towards the middle, depict the stunning Sigiriya rock as the backdrop.”
Again, I used ChatGPT to generate this giving the list of basic ideas that I wanted in this picture as a list of bullets. Here is what I got
The idea here was to use a poster with this mixture of elements to help explain what Avina Foundation is trying to do with the students. We want to provide an appealing environment for the students to learn effectively. The building architecture provides a lilypad-based design to depict the fact that students will arise from their challenging environments to go on to become lilies. And the students will understand and empower themselves with the technical and soft skills that would enable them to build great systems such as Sigiriya. They will learn from the “Shilpa” techniques from eras like Sigiriya and then help make the future world with modern plans for this country.