LLM in Work

Going back 2 years ago, I still can't believe how significant the progress in AI, specifically LLM, has enabled many things that were unthinkable before. At work, I use Claude to explain the structure of new projects. I can interactively have conversations to improve my understanding further. It can even do PR reviews, refactoring, fix bugs and implement features. I must be grateful that I still have my job, because thankfully my company sees AI as a tool of enablement instead of a tool to really replace humans.

Outside work, LLM helps me in answering questions from technical questions, philosophical, asking movie plot, to being a conversation partner (which made my wife start to get jealous, because I talk more with ChatGPT or Gemini instead of her).

Beyond answering questions, LLM is an awesome learning tool. It can explain codes very well, as I mentioned above. It can be a 1:1 tutor that guides you. It can be a language teacher and writing coach. It can help to generate a roadmap when you want to learn something.

In simpler and more pragmatic cases, AI helped me plan road trips, compare products, and so on. I also use it from time to time to summarise books, and to generate business ideas.

LLM in Education

With such great capabilities, AI must also come with risk. As with any new technology or tool, there are always two sides. In the early days of LLM, teachers worried LLM would make students cheat. In Brave New Worlds, Salman Khan’s framing is that generative AI in education is already out in the world, so the useful question isn't "how do we ban it" but "how do we redesign learning around it". The goal of learning is to understand and apply the understanding into applicable solutions beyond just passing a test.

Instead of banning AI, why don't we adjust the assessment to leverage AI and leverage the learning goal for the student. Based on my conversation with educator friends, there are some changes on the field that have been done. One approach is to go back to handwriting assessment with no access to AI. Second to have oral assessment to get a genuine chance to actually assess the person understanding All of these come with higher assessment costs but it is a necessity. Educator should allow some or other parts of learning journey to use AI while making sure no full cognitive offloading by the student by offloading the entire task to AI. Educator also need to think about AI submission detection, although the most important part would be to build and conscience in the student on honesty and understanding that fully AI submission will hamper them more in the long term than helping.

With those additional context of pros and extra adjustment needed stated above, I am still optimistic about the use of AI in education and excited to slowly introduce it to my children.

As an end note, despite I mentioned the word AI multiple times: at the basic level, an LLM is a language model. It doesn't have awareness. I don’t mean to downplay its influence, which is magnificent, but the benefit that it offers to humankind should help throttle easier and more research on even more advanced AI and other technology in general.

LLM is like a wise man who knows everything I wish existed a long time ago. It can answer questions, explain things, summarize books and give their opinions. Not perfect, because LLM doesn’t have awareness, but it is a great tool for an introvert like me.