The post below accompanies my 4-Quarter Lives podcast interview with fellow INSEAD alumnus, Graham Heddle. He’s been using and teaching how to harness the power of AI for personal and professional transitions after finding it useful for his own shift after a quarter century career in private equity. It’s fascinating and unexpectedly powerful. And I don’t think most people are yet using its power for personal transformations. Give it a try. Guaranteed to blow you away… And Graham has kindly contributed this short guide below to get you started.
The Context
For the history of AI, Ray Kurzweil, of Kurzweil synthesisers and much more besides, wrote this well-known book. It's actually 18 years old now but still relevant for background and principles: The Singularity Is Near. He is now principal researcher and AI visionary at Google Mind. He also features in the following film on the same subject, including a tv clip of him demonstrating a piece of music written by a computer (one of the first) in what looks to be early 60's to me. YouTube: Transcendent Man. For a current music writing AI go to suno- ai.org.
Homo Deus by Yuval Noah Harari is a brilliant description of why our human future might be dystopian. In all of my research, talking to many people, I would still say the views are 25%:25% diametrically opposed. This tells me the world is equally divided amongst natural optimists and natural pessimists with the other 50% somewhere in the middle. There is no- one, almost literally no-one, who isn’t at least somewhat concerned, however. Usually, the concerns run along the lines of this: it’s all happening too quickly and the regulators aren’t keeping up. It will probably be ok-ish in the West, but what will the Chinese and North Koreans and other untrustworthy people in the Middle East do with it? What does it mean for society, work and welfare? What does it mean for my children? Hell, what does it mean for me?
Reading Prep
For a more sober, and highly sensible view of what the practical implications and likely economic effects will be, I still haven’t read anything better than this book written 5 years ago by the leading UK economist Roger Bootle: The Al Economy: Work, Wealth and Welfare in The Robot Age.
You should definitely read this book by Wharton Professor Ethan Mollick, which came out in March: Co-Intelligence: Living and Working with AI. Very approachable and well-written. I have spoken to both him and Henning Piezunka at INSEAD who arranged the following webinar in his excellent “Between The Lines” series. The webinar is now available publicly with the same name here: Co-Intelligence. If nothing else, watch minutes 24 to 30 and 58 to 60. His Substack is well-worth subscribing to and you can do so for free here: One Useful Thing.
On the enormous questions of how the brain works and consciousness arises, and how all of this is related to the development of AI, there are two great books:
David Eagleman: Incognito: The Secret Lives of the Brain
Max Bennett: A Brief History of Intelligence: Why the Evolution of Intelligence Holds the Key to the Future of AI
For an insight into the latter, have a look at Part I (only 9 mins) of this webinar series with Max Bennett here. I have also produced a very short description of the difference between transformers (eg. Large Language Models “LLMs” like ChatGPT, Claude, Genesis and Llama) and the human brain.
How to use level-4 models (eg. Chat GPT-4o etc)
Since mid-May ChatGPT-4o (‘o’ stands for “omni”) is available for free through openai.com. This replaces ChatGPT-3.5 and is an order of magnitude better, in that it doesn’t make the really silly errors of GPT-3.5, is much more powerful, is multi-modal (can deal with text, voice, image and video) and you can upload documents, spreadsheet, presentations etc for it to work with.
Prompting
There is a load of info on the web about prompting. Lots of it is out of date because it relates to the now basically defunct GPT-3.5. Some people who had learnt quite early on how to phrase questions to elicit better responses from ChatGPT (particularly in the US) started calling themselves “AI Whisperers”. When you have finished vomiting, disregard them
because a) things have moved on, b) you really need trial and error yourself and c) the coming versions of all the LLMs will be WAY better at interpreting a user’s questions and framing them for themselves to obtain better prompts, and therefore more compelling responses to your questions, so eventually you won’t need to know much about prompting at all. But you definitely do for now.
If you get stuck when asking questions to the LLM, try using the methods outlined below (and more advanced ones) to generate prompts by asking the LLM itself how to question it (yes, really: “what prompts do you suggest I use to help you address my question” is useful). This type of question can lead to interesting discussions with the AI - believe me, I’ve been up all night more than once, so fascinated by where the discussion has gone, that I ended up not sleeping at all. Or as Ethan Mollick puts it and to paraphrase him: “until you’ve had at least 3 sleepless nights you haven’t played with AI nearly enough. You are going to be both amazed and completely freaked out when you do”.
Examples of Prompts
Generic Prompt
Question: "How many hours are there in a week?"
Prompt: Enter the above as the prompt.
Chain of Thought (CoT) Prompt
Question: "How many hours are there in a week?"
Prompt: "To calculate the number of hours in a week, start by remembering there are 24 hours in a day. Since a week has 7 days, multiply 7 days by 24 hours per day to find the total hours in a week."
Comparison
Generic Prompt: The question is posed directly without any hints about the steps needed to find the answer.
Chain of Thought Prompt: This prompt includes both the question and a detailed thought process that outlines the steps to reach the answer. It models reasoning that leads logically to the solution.
CoT prompts help define the approach to answering the question whilst at the same time clarifying for the user (and the LLM) on how to think about the problem systematically. This method is particularly useful in educational settings or when trying to solve complex problems where the steps are not immediately obvious. Trial and error will show you how to cast your prompts. This will gradually become intuitive until you can just “tell” whether a prompt is likely to be successful. If the LLM leads you down a blind alley, ask it to stop so you can start again. If there’s a long response coming (you can tell by the numbering and number of bullet points), which already looks misinformed, click the black button to stop the LLM responding and try again.
Prompt instructions – ask an expert
Rather than: “I have a wine export business, how should I grow sales?”
Try asking this: “You are a leading wine exporter with many years of experience. You have consulted widely for other exporters and helped them to grow both their sales and the end market for foreign wines. You focus on the US and are used to dealing with multiple state laws and regulations. How would you advise a new client to grow its market share? The new client has a low volume of sales in just 2 states currently”.
For further ideas on prompting: try this article and see a longer example below.
Graham Heddle – London, May 2024
A Brief Guide to Prompting
A simple overarching rule:
Example of how to prompt for a "Devil's Advocate" view of an idea:
Prompt 1: “You are the devil’s advocate, and you challenge my idea Only provide Advantages and Disadvantages; only use the information provided"
Example of an interesting prompt to optimise a LinkedIn profile "About" segment:
Copy and paste current LinkedIn Headline and About section into AI (or attach the pdf of your profile). Then prompt the AI:
"You are a personal branding expert, skilled at crafting LinkedIn profiles. The following [attached] is a LinkedIn headline and "About" section for [insert your name]. Your goal is to improve the headline [or the "about" section]. Please suggest 3 alternative LinkedIn headlines of no more than 220 characters each that are both memorable and keyword rich"
For further prompts on prompting: try this article.
CoT Prompt Detailed Example – AI and Blockchain
Simple Prompt:
"Can you explain how blockchain technology can be leveraged to improve supply chain transparency?"
CoT Prompts:
1. Introduction to Blockchain:
• "Explain what blockchain is, focusing on its key features like decentralisation, immutability, and transparency."
2. Identify the Current Problem in Supply Chains:
• "Start by defining common issues in supply chains today, such as lack of transparency, risks of fraud, and inefficiency in tracking products."
3. Application of Blockchain to Supply Chains:
• "Discuss how these features of blockchain could address specific problems in supply chains, such as improving the accuracy of asset tracking and reducing unauthorised tampering."
4. Benefits of Blockchain in Supply Chains:
• "Enumerate potential benefits such as enhanced transparency, increased efficiency, and reduced costs associated with supply chain management."
5. Challenges and Considerations:
• "Analyse potential challenges that might arise when integrating blockchain technology into existing supply chain systems, including technological adoption barriers, the need for standardisation, and regulatory considerations."
6. Conclude with Practical Examples:
• "Provide examples of companies or industries that have successfully implemented blockchain technology to enhance transparency in their supply chains."