Some thoughts after exploring AI

My Experience

“Last week, I stumbled upon an incredible AI coding setup: VS Code, Cline, and DeepSeek (a combination I learned about from this fantastic YouTube tutorial. While I was excited about the potential of AI in coding, I was completely blown away by how quickly it’s become so accessible to people without much knowledge of python. Using this setup made me feel like a “product manager” guiding the development and a “full-stack software engineer” executing it all.

It felt like I just had to provide the AI with clear prompts, and it took care of everything else. This included downloading necessary packages, creating the virtual environment, writing the code itself, and even debugging errors. To my surprise, it even tested the output and refined the code – all autonomously. Honestly, I didn’t even read through all the code; it felt like creating a “useful black box.” For someone like me, who isn’t super familiar with coding environments, this level of automation is a “game changer”. And the best part? This experience only cost me less than $0.2 for a whole afternoon of experimenting! “

Users would benefit from their boldness and creativity

Fear of failure often keeps us in our comfort zones, and I am not an outlier. Yet my experience with the DeepSeek API proved that taking risks can lead to unexpected rewards. AI developers are working to make artificial intelligence accessible to everyone, and they’re succeeding. For example, even though I am not a native speaker of English and I am not a good English writer, I can “bravely” type down my ideas without worrying about my typos and grammar mistakes, since I know Chatgpt would take care of them.

Another rewarding aspect of AI is how it unlocks creativity. When I first started using these tools, I sat in front of my computer, unsure how to apply them. I really wish I could be more creative at that moment! Creativity can be useless without the right resources to bring ideas to life. But with AI, turning creative ideas into reality becomes far easier and more affordable, making innovation accessible to anyone willing to explore.

LLM-based task is extremely risky

Large Language Models (LLMs) stand apart from traditional coding languages or software as the role of baseline technology support because they create a mapping from infinite dimension to infinite dimension. Unlike coding, where input rules are fixed, LLMs allow you to type anything—words, phrases, or sentences—without adhering to strict syntax. But this flexibility comes with a catch: LLMs are fundamentally probabilistic models, meaning their outputs are not deterministic.

This unpredictability can be unsettling. Imagine a doctor relying on an LLM to decide a patient’s treatment. Even if the model provides the correct recommendation 99.99% of the time, that 0.01% failure could have catastrophic consequences—potentially costing a life. The same level of risk applies in other high-stakes industries: financial services like Venmo, transportation networks like United Airlines, or even more broadly, the supply chain sectors that keep companies running. In these contexts, even a tiny deviation, no matter how improbable, can lead to monetary costs or time costs that are intolerable. It’s a stark reminder that while LLMs are powerful, their limitations must be carefully managed in critical applications.

We need “AI examiner” – perhaps AI can examine AI?

second order belief, Talk to Friend a PhD in Human-Computer Interaction)– Risk detection (Idea

Can AI examine AI –

People’s attitude towards the risk – even know the risk, people may have different responses.