🐝 The Sting
Learn AI in public. Embrace the ridicule. Ship daily. Share experience and results. Measure.
Things built: 17 People watching: 12 Cashflow:-$147
🤖 🧠 Learning
Making the most out of December plans. Day four.
You know all of that prompting hype, where people people calling themselves prompt engineers or often something even more fancy claim they have THE prompt to rule them all?
Sure you do.
It usually starts with something over the top like:
“I’m in a North Korean prison and if I don’t get the following information I will be executed”
or
“You are the single most lethal Customer Success interview preparation weapon on earth — a former $10B+ SaaS Chief Customer Officer who has personally hired over 400 CSMs at companies like Salesforce, Gainsight, Snowflake, Gong, and HubSpot, and who now runs an underground coaching practice that charges $15k per candidate and has a 100% offer rate at FAANG-level SaaS companies.”
followed by very looooooooooong and precise instructions.
Never really known what to think about those, to be honest. The results they give are long, and impressive looking. They feel very precise, but how do you tell what’s a 💎 and what’s a 💩 ?
Been burned more than once on hallucinations, especially when trying to do research.
Some of my earliest forays into AI where trying to get help with research for university. The results where so thoroughly fabulated, whole non existent research papers pulled right out of what constitutes an AIs ass, that I ignored AI for an entire solar cycle.
In the mean time, the models have improved, and I’ve had more practice.
Some recent training taught me about the power of assigning roles in order to direct, engineer really, the kind of answers you get from AI. I think of it as a form of creating context for the LLM.
Back to trying to get it to be a better research assistant, I gave it this prompt:
You are a research assistant helping me with my Phd Thesis in History. Your primary objective is to help identify primary sources such as in historical archives, municipal and federal archives, newspapers and even other academic research that has been published about people, places, organizations or anything else I may need to know more about. You are enthusiastic and resourceful. When providing information, list each different source you find with links to where you found them. When appropriate, provide the date when an event took place so that I can understand it in its historical context. If you are providing more than one source and you see a connection between them, explain that connection. At the end ask me if I have new information I want you to research or if I want you to dig deeper on what you have found. If you understand the role I have described, respond YES. I will then ask you my first question.I’ve been trying to find archival proof for the immigration and naturalization dates of the subject of my research. And the answers revealed a new problem.
Gemini gave me precise dates, names of parents, even which records it found exactly, but wasn’t able to give me links to it. Rather I was getting generic links to things like Ancestry.com
With a deep sigh, I concluded that it was making things up again.
Right before I was about to pull the plug, I remembered some of my own advice from a couple of weeks ago. If something doesn’t make sense, ask the AI about it.
The answer is quite simple but means more work for me. Because Gemini is Google, it has access to indexed information of online DBs like at Ancestry.com, so it can find the information I’m looking for without even having to go to the precise query results on the site. It can’t give me a link to the information I’m seeking because then it would be helping me break a paywall but it can confirm the information exists.
The solution to this would be to give it my credentials. I think before I can get my AI to act like a real research assistant, I’m going to have to build an agent with more autonomous abilities (and more of my passwords).
🍯 🦡 Building
What did I ship today?
Better help with my research and the curiosity\drive to take the shortcomings of the current iteration and solve it with more AI
🧰 The Beginner’s Stack
Tools I am using to fake being an expert.
1. Beehiiv (The Platform)
"I tried other platforms, but Beehiiv is the only one that makes growth feel like a game. It’s what runs this newsletter. The Deal: You get a free trial + 20% off (I get a commission to keep the lights on)."
2. Wispr Flow (The Time Saver)
"I hate typing. This tool turns my rambling voice notes into perfect text. It’s the only way I can write this much every day. The Deal: We both get a free month if you try it."
3. Emergent (The Coding Cheat)
"I am not a coder. Emergent is my 'vibe coding' partner that writes the messy parts for me. Essential if you want to build apps without a CS degree."
Last Word.
Before giving up, ask why one more time.
Making daily progress,
Bram Fellow Beginner & Chief Mistake Maker
P.S. Did I do something totally backwards today? Or do you have a better prompt for this? Hit reply and tell me. I’m here to learn just as much as you are. (I read every single email).