Lessons from Two Months in AI

Lessons from Two Months in AI

As a child, I remember my parents telling me that the test of true intelligence is the ability to hold two opposing ideas at the same time. While it turns out that was F. Scott Fitzgerald, and not a Benn family original, I’ve thought of it often as I’ve dived deeper into the world of AI, a significant change from my pre-MBA life in international development. Going from seven years in one field, to an MBA, to navigating the uncertainty of what comes next, has been an equally exhilarating and exhausting journey, and one that merits its own post.

In April, I had the privilege of joining Pioneer, an early-stage start-up led by warm, supportive co-founders who have encouraged me to take time each week to learn about the explosive world of AI & GPT-4. At this stage, I find AI both terrifying and exciting. It has the potential to generate harmful content at an unprecedented scale (think fascism bots or misinformation machines), yet it also has the power to make our lives exceptionally easier (join me down the Fully Automated Luxury Communism rabbit hole, my hopeless optimists).

Over the past two months, I’ve kept a log of my constantly changing thoughts, opinions, and lessons learned. Below are just a digestible few, for those who need to start learning AI as we rocket into the unknown future (hint: every person reading this now).

You should care.

Yes, it’s very cool to blow off AI as trendy, or roll your eyes at GPT-4, or scoff that AI isn’t relevant because your field is in XYZ underwater basket weaving. But if you want to stay ahead of the curve, or even keep your head above water, it’s time to learn and prepare yourself for an exponential rate of change in the foreseeable future (not just AI, but the climate crisis too. We’re talking extreme shake-up of both our physical and virtual landscapes).

There’s an ongoing debate of whether AI will replace your job or whether a person using AI will replace you. As of today, this is basically a philosophical argument. Instead of grinding your gears fearing job displacement, redirect your energy to learn prompt engineering, play with various APIs, and take the time to turn yourself from good to great, or great to exceptional.

Ask yourself: what’re the most tedious and intellectually unfulfilling tasks in your work? Can you automate them, freeing up your time for more meaningful and creative endeavors? AI is a tool that can enhance your capabilities, making you more efficient and allowing you to focus on the aspects of your work that truly matter. And it’s more than just GPT-4 (or Google Bard or Microsoft Bing).

Tired of taking meeting notes? Try Zoom Read.

Need to triage your inbox? Try Uify.

Having trouble digesting that 200-page report? Try Any Summary.

AI is not magic.

GPT-4, and other LLMs (large-language models) are not mystical forces that generate ideas out of thin air; they are powerful instruments that require our understanding and expertise to wield them effectively (….for now. Scifi buffs, optimistic futurists, and paperclip maximizers, DM me and let’s talk AGI).

They require precise instructions and guidance from humans to operate effectively. Some useful tips for GPT-4 beginners:

  • Shorter isn’t better. More detailed prompts with clear commands (write, classify, summarize, etc.) will give you better outcomes.
  • Create prompts that encourage step-by-step thinking: “explain your step by step process” or “lets think step by step”. This is called “chain-of-thought”.
  • Stuck at a dead end? Adopt the Socratic method to stimulate brainstorming and generate better questions. Try the prompt below next time you’re stuck and are looking for a virtual sounding board:
  • Ask GPT-4 to challenge your assumptions and provoke further thought. What’s the other side of the argument?
  • ELI5 (Explain Like I'm 5) difficult topics, articles, or papers to gain a deeper understanding.
  • LLMs are prone to recency bias (wild). If you’re copy & pasting a lot of text, put your instructions at the end of the prompt.
  • It always takes iteration. First prompt kinda sucks? Keep trying.

Understanding the harms.

GPT-4 isn’t always right, and when it’s wrong, it’s confidently wrong. A few weeks ago, I tried to develop a prompt to have GPT-4 pull a summary from my LinkedIn profile, and it told me that I was the CEO of a blockchain start-up that graduated from Harvard. Not so much. This is called hallucinating, and is why you should always seek sources, check those sources, and interrogate what it’s giving you. While always improving, it’s not foolproof.

Clever people can also trick LLMs through jailbreaking. My favorite:


It’s crucial to acknowledge the potential for harm within the current AI ecosystem. Calls for a six-month pause on AI development have been made from leading global experts, but the reality is that the AI genie is already out of the bottle, and the market is charging forward rapidly, without guardrails, to the detriment of us all. How will we value the reduced need for human labor? Can we significantly increase global productivity without widening inequality and reducing the tax bases?

While regulation is vital, we know from snail-paced social media & big tech regulation that the government struggles to keep pace with lightning-fast tech advancements as a rule, not an exception (hello teens, social media, and the mental health crisis).

To shape a responsible AI landscape, we must address these flaws and work towards ethical practices and regulations. Here are seven excellent policy recommendations to combat risks from advanced AI systems. Not convinced? The Effective Altruists (shout-out to the workshops I took while in Oxford) explain AGI alignment & safety the best.

Preparing for the future.

AI is an undeniable force shaping our present and future. You can recognize the potential for harm and understand that AI is a powerful tool that we can navigate with purpose. AI can supplement you, but it shouldn’t replace your wit, your creativity, or your lived experiences. It can make your writing and thinking faster, and the tasks you despise more bearable.

In the last 20 years of social media & the internet, noise has increased to an almost unbearable amount. With AI, people will outsource their knowledge work, creating an avalanche of undifferentiated, lazy, boring content. Keep your wit & your unique view, but use AI to help. Encapsulated perfectly below:


For those who want to dive in deeper:

Finally, to my skeptics, including myself a year ago who would not believe the turns my career has taken… playing with LLMs is fun. And impressive. We’re at the industrial revolution moment, the internet moment, the green revolution moment. You can be an early adopter, it’s not too late. How cool is that?