LLM, Value, and Fruits
Since I signed up for my openAI account and typed my first prompt to chatGPT, I’ve been using it almost every day. ChatGPT can do so many different text generation tasks [1]. My favorite feature is rephrasing (to fix grammar errors and be more fluent or official). The results are impressive. I’ve also tried knowledge search and suggestion a few times. The outcomes were impressive but not useful. They were hand-wavy or too high-level for my needs. Most importantly, the generated texts didn’t provide a direct connection to my specific task.
The rephrasing feature of generative AIs is super useful. I start with a rough draft, in which I include all the important information without worrying too much about grammar and flow. Then rewrite it to be grammar-error free, and more fluent or more official. Quillbot has a neat Chrome extension that is so tightly integrated into my writing workflow in the web browser. And then I often rewrite it again to convey the exact meaning and intent. It’s butter smooth experience. It's an ideal tool for my writing routine, where completing okay-quality work in a short amount of time matters.
The interactive collaboration between me and generative AI tools has seemed perfect in any writing, until I used it in a different use case. Recently, I wrote a recommendation letter for my friend. He is a junior-level SDE who now pursues a data scientist career, starting with the Masters program. I wanted to highlight his strengths of fast learning and analytic thinking. I started writing the letter as I used to do. After a couple of iterations, I found how hard it was to both highlight key messages and align each with the actual facts in the episodes. This writing includes complexity, genuineness, and uniqueness. I was bothered a lot by factual changes and blurred key messages by the AI tool. After a few attempts of trial and error, I decided to write from scratch by myself. I still got a basic grammar check using the tool on my very first draft, but I double-checked (probably triple-checked) it myself before submitting it.
Where do you want to apply generative AI withtout being bothered by its quality, and where you don’t? It’s a good topic to discuss with people from many different backgrounds, including AI science, product development, consumer studies, policy, art, and humanity. Depending on the applications, we may need different technologies than LLMs. Every technology has its own low-hanging fruits and high-hanging fruits, and LLMs are nothing but predicting the next word, trained on a large scale accessible past data (and human feedback).
References
[1] https://medium.com/mlearning-ai/the-chatgpt-list-of-lists-a-collection-of-1500-useful-mind-blowing-and-strange-use-cases-8b14c35eb
