
Hacking Academia
An ever-growing series of tutorials (with detailed notes) filled with practical, experience-driven tips and tricks for being effective, happy and successful in modern day academia and related careers.
Hacking Academia
Experiences With and Thoughts on Artificial Intelligence: April 2025 Update
Time for an update on #ArtificialIntelligence, from my personal perspective in a professional context: my continued exploration of its intense usage in a day-to-day career (2 and a bit years and counting).
Key concepts include:
đź§ Force Multiplying Experts More Than Beginners
In my experience, these AI tools are helping experts more than anyone else. If three things come together:
🔹 an expert who knows very well how to do the task entirely themselves
🔹 someone experienced at giving detailed, nuanced, and error-robust instructions to achieve an overall goal and sub-goals
🔹 someone who can effortlessly glance at output and judge whether it’s correct (again, they need to be an expert in the task they're using AI to assist with)
✨ Then these tools are like adding a superpower – but their benefit diminishes if even one of those components is missing.
🛠️ There’s a good and bad way to integrate AI into your workflow: forcing it through a rigid list of steps is risky. The best results come from flexibly exploring where it can help most and adapting around its shortcomings.
⚡ They’re great at helping maintain momentum when you’re tired, sick, unmotivated, or just not feeling it. It takes a lot less to “get over the hump” with AI assistance than when it’s all on you.
🔄 I can iterate and work with these tools guilt-free – no etiquette, no need to provide fulfilling work like I would with a colleague. That freedom to rapidly try and discard half-baked ideas without wasting someone’s time is huge.
(Yes, I know some would say that’s already how some profs treat students, but I hope that’s the exception, not the rule.)
đźš§ Key challenges that remain:
🧩 Knowing how much to pre-prepare and modularize a task – especially as newer tools reduce that need
đź§Ş Knowing whether repeated failures are from bad prompting or fundamental limitations of the tool
🛑 Knowing when to go manual-only – particularly at the polishing stage – and accepting that this varies project to project
đź”® I also share some thoughts on where I see all this heading, based on 2.5 years in a post-ChatGPT world.
đź’» YouTube link: https://youtu.be/39Sz5ipFoqg
#artificialintelligence #AI #technology #disruption #prediction #chatgpt #OpenAI #Anthropic #MidJourney #AdobeFirefly #Firefly #Suno