Fifteen years ago. PhD candidates conduct research for a long time. I’m overwhelmed by many articles, notes, emails, bookmarks, etc. I felt very relaxed when I found a reference manager tool Mendeley. It’s like I’ve controlled the process again. When I found the bookmark manager Xbookmark, I felt very productive (still bookmark). They worked very well for me at the time and I completed my PhD program and earned my degree.
Episode 1 – Face Reality
These days, I have an existential crisis. I’ve seen advances in AI research assistant tools. I worked with Scinito a few days ago and was shocked. I tried to convince myself that this was just a tool to help with literature reviews. Those who earn their PhDs know how difficult it is to have a solid literature review. This is not a joke. You have to read over 100 articles, classify them, understand them and summarize them. If I say I did a literary review 3-6 months ago 15 years ago, I’m not wrong. You heard my precious life for 3-6 months, and you heard my voice.
At first, I tried to convince myself that Scinito or other similar tools provided marginal value to researchers. Sadly, or very happy, I was wrong…
Not only can they make literary comments for you in a minute (sorry, my fellow researcher, you heard the right news), but they can read your article carefully. I will never forget that I should wait for my mentors and consultants to review my posts, how many cover emails we have until things become acceptable quality. Even after all these efforts, you will get extensive feedback from peer reviewers in the journal to consider your article for publication. Or, it was rejected for 3 or 6 months simply because of choosing the wrong diary to publish. These AI research assistant tools can enhance all these steps: View your articles and select the most relevant diary for you.
Amazing. It’s really amazing for researchers of this era, but I’m sorry to see how much time I spent on things that could have been easier and faster. Interestingly, this is not the end of the season. This is just the beginning.
This challenge is not only for researchers. It also has to do with software developers. Tools like the Cursor IDE have changed the way we build software in a huge way. After my PhD, I started my career in engineering. So, I did a lot of coding, testing, etc. Today, I don’t need to read Stack Overflow to debug my code. I don’t need to spend time writing tests for my code. I no longer need to be an expert in React, HTML or CSS to build a website. How much time did I spend on construction sites in the past? I don’t want to think about it!
Episode 2 – Embrace Reality
Let me share half of the glass experience too. This is super cool. I can ask the AI Research Assistant tool to perform semantic searches in an extensive database. Things we couldn’t do before. It’s just a keyword matching. By reading a literature review generated by AI in seconds, I can get the latest information on any topic or research question in a matter of hours. I can easily write latex code. I can reformat the paper in a few minutes. I am happy for the researchers of our time. They can spend more time on creativity, problem solving, and of course valuable life, rather than doing unnecessary time tasks.
I am happy for myself, too. I can write code in any programming language I want. I can build a website without locking in WIX or WordPress. I can write any Python code I need. I can optimize it and write a series of tests for it. Wow! This is so cool. The landscape of coding, design, research and everything is developing rapidly. No matter how many people or organizations resist, technology will find its own way.
There is a catch here. The promise to build a (only one) prompted website is incorrect. I’m telling this based on my current experience. Today, I am working on a new website with my colleagues, and we are both experts in software and AI. We didn’t even think about Wix or WordPress this time. We started using Culosr and its proxy experience on Claude-3.7-Sonnet. The cursor’s proxy can generate the website structure in one second, but fail in details.
For example, when you want to align two different texts with each other, especially when one is static and one is dynamic, AI can’t do this correctly. Basically, AI can complete the website structure in one second, but can’t execute the required details (you want to be a detailed information for human applications on top of the prebuilt structure) as well as UI design experts. This means that even if we don’t need to be experts in React or CSS, we must know the basics of intervening in the code base when needed. Also, we have to understand these concepts well to elaborate on them. If you can’t say that, AI can’t do it!
This weakness of the AI model is not shocked. They are based on the principle of “wisdom of the crowd”. This means they are based on the most common aggregates, rather than mimicking the intuition of an individual expert. This stems from their fundamentals. Their universality is amazing, but they suffer in particularity. In this brief podcast, I explain a similar concept from different perspectives: “specific erosion”.
Last words
I’m lucky to be a part of the AI community. I am an AI architect with a solid plan to accept this technological change. But I am worried about many people who cannot manage this change. This is not easy at all. If you have a moment of survival in your career due to AI, please let me know. I probably know something helpful.
If I could share a tip here, it is “A deeper understanding of the basics.” You can (should) leave repetitive, advanced and universal tasks to AI and use your human creativity and expertise to detail to make your work/product shine.
If you want to hear more stories from the perspective of AI architects: youtube.com/@aiunorthodoxlesson/, please follow me on YouTube
Awesome https://shorturl.at/2breu