🐙 Promptpack: Generative AI for exponentialists
Prompts to help you make sense of the present and understand the future like an exponentialist.
Hi,
here.At Exponential View, we have embraced generative AI tools since they were opened to the public last year. Over several months, we’ve explored prompts, diligently tested a range of tools, and created an environment of sharing and learning amongst team members. Throughout this experimentation, we’ve started to see how genAI best helps us do what we are known for: analysis of the exponential future.
I wanted to share with our readers a first Promptpack, an easy-to-use series of prompts that have drastically helped me make sense of trends and events. Based on the nature of our work and LLMs’ current strengths and weaknesses, we use ChatGPT in an exploratory way, as a thought partner rather than a search engine or to receive single, truthful answers.
In this post, I’ll share prompts based on studies from the emerging field of prompt engineering. Zero-shot prompting (one prompt without providing examples, as opposed to few-shot prompting) is the technique we will focus on here, as I have found this particularly effective for rapidly gaining understanding on a given topic.
TL;DR
Prompting today is more art than science, and open-minded experimentation is the way we approach it.
Generative AI as a thought partner can help you find unlikely connections between ideas, apply concepts and theories to new situations, explore future scenarios, and question your assumptions about the future.
This post will equip you with four exceptional tried-and-tested prompts which can help you in brainstorming, strategic, analytic or planning work. It’ll turn your LLM of choice into a powerful (and pretty reliable) coworker. I’ve also thrown in two bonus prompts, which I think you’ll love.
But first, prompt engineering: art or science?
Large language models use neural networks to learn from a lot of data. Then, based on user input, they output what the model perceived as a relevant response. While large language models like GPT-4 generate text based on probabilities learned from training data, the output isn’t deterministic for a given prompt. Instead, it includes a level of randomness, often controlled by a parameter known as “temperature”.
When you provide the same prompt to a language model multiple times, it’s like rolling a weighted die each time. The weights (probabilities) haven’t changed, but because there’s an element of randomness, you can still get different results.
This means that ‘how’ you prompt matters, as the model may choose different options depending on the wording and its associations. It also means that the output will always include some element of randomness. The more outputs you get, the more ground they will cover. This is an important principle that I use: ask for multiple options, so that I can decide what is most useful for me, and I can see a range of possibilities.
Another reason to ask for multiple answers is the evidence around chain of thought (CoT) prompting. CoT is a technique that has been shown to drastically improve output by encouraging a chatbot to explain its reasoning. Further testing has found that saying “Let’s work this out in a step by step way to be sure we have the right answer” yields the best zero-shot prompt results. I incorporate this into my prompts, or ask for multiple results, depending on my needs.
At this point, prompting is more art than science. This is especially true for more subjective research, which is difficult to put to the test; testing prompts requires falsification against benchmarks. In other words, the outputs resulting from the prompt need to be compared in an objective way to outputs from other prompts, or to work that humans may “output”. When the result is very subjective, it becomes more difficult to compare the quality.
Read more about the emerging science of prompting in the following articles:
Hebenstreit, Konstantin, et al. “An automatically discovered chain-of-thought prompt generalizes to novel models and datasets.” arXiv preprint arXiv:2305.02897 (2023).
Coda-Forno, Julian, et al. “Inducing anxiety in large language models increases exploration and bias.” arXiv preprint arXiv:2304.11111 (2023).
Zhou, Yongchao, et al. “Large language models are human-level prompt engineers.” arXiv preprint arXiv:2211.01910 (2022).
Also read EV member Marshall Kirkpatrick’s incredibly useful guide to prompting for strategic thought.
Now, let’s get practical!
Prompt 1: Find connections
The world is increasingly interconnected and complex. AI can help us find the hidden, unlikely connections between things. When thinking about exponential technologies, it’s useful to see which are strategically important for what, and how they could enable other things. I’ve found this to be especially useful when thinking about geopolitics and materials.
For example, I have an inkling that lasers could become an increasingly important part of the future. Is that true? How do I ask the question to ChatGPT while avoiding it hallucinating to satisfy my desire to be correct?
CS: Is there a relationship between lasers and geopolitics? If so, give me 7 examples.ChatGPT: While it might not seem immediately apparent, there is indeed a relationship between lasers and geopolitics, particularly in the realms of military technology, energy production, communication, and space exploration. Below are seven examples demonstrating this relationship:
- Military Applications: Lasers are used in a variety of military applications, such as target acquisition, countermeasures, and even as direct energy weapons. The geopolitics here lies in the balance of power, as countries with advanced laser weaponry may have an upper hand in conflicts or during negotiations. The development and deployment of such technology often lead to arms races.
- Anti-Satellite Weapons: Lasers can be used to blind or damage satellites, which is of great concern in the context of geopolitics since satellites are essential for communications, navigation, and surveillance. A country's capacity to harm or disable another's satellites can drastically affect the power dynamics between nations.
- Laser Nuclear Fusion: If sustainable fusion power becomes a reality, countries with advanced laser fusion technology could potentially control a powerful and nearly inexhaustible energy source. This would significantly influence geopolitics, as current global power structures are largely influenced by control over energy resources.
- Telecommunications: Fiber-optic cables, which use lasers to transmit information, are an essential part of the global communication infrastructure. The control over, or sabotage of, these cables can greatly influence global connectivity and information flow, thus becoming a major geopolitical issue.
- Space Debris Management: With the increasing issue of space debris, lasers have been proposed as a solution for deorbiting smaller pieces of debris. Countries or organizations that control this technology could exert influence over space traffic, which is becoming increasingly important as space travel commercializes.
- Laser-Induced Earthquakes: There's research into the ability to trigger small earthquakes using high-powered lasers. While this technology is still speculative and in early stages, if it comes to fruition, it could have significant geopolitical implications, potentially leading to new types of warfare or sabotage.
- Underwater Communication: Lasers can be used for underwater communication where traditional radio waves don't perform well. This technology can affect geopolitics in terms of naval warfare, submarine communications, and oceanographic studies.
These examples highlight the fact that while lasers are a specific technology, they can influence geopolitics in a number of diverse and powerful ways. As laser technology continues to evolve, its impact on the geopolitical landscape will likely grow and change.🔮 Copy this prompt: Is there a relationship between X and Y? If so, give me 7 examples.
Prompt 2: Apply frameworks
Added context helps the AI become more creative, and give answers that may be less obvious to most humans. Giving ChatGPT a theory to apply serves the double purpose of helping us better understand something through a familiar theory, and enabling the AI to provide better results through the added context. For example, at Exponential View we understand technology as being driven by learning curves and modularity. Finding these across the themes that we research can be a powerful signal of coming innovation.
CS: Apply the concept of learning curves to solar panels. Give me 7 options. Let’s take this step by step to make sure we have the correct answer.ChatGPT: Sure, the concept of learning curves is generally used in the context of manufacturing, and it represents the idea that as production increases, the company will learn to do things more efficiently and effectively. In other words, the more a company produces, the better it gets at it, thereby reducing the cost of production per unit. The cost per unit is known as the "experience curve effect."
Applying this concept to solar panels, we might expect that as more solar panels are produced, the process would become more efficient, leading to a decrease in costs. Let's look at several ways this could happen:
- Increased Production Efficiency: As the volume of solar panel production increases, manufacturers are likely to gain experience and improve their techniques. This will lead to reduced waste, improved time management, and better utilization of resources, all of which can reduce the overall cost of production.
- Technological Advancements: As more solar panels are produced, there may be increased investment in research and development, leading to new technologies that can make the production process more efficient or the panels themselves more effective. These advancements could make the panels cheaper to produce, or they could increase the energy output from each panel, effectively reducing the cost per unit of energy produced.
- Economies of Scale: As the production volume increases, solar panel manufacturers can purchase raw materials in bulk, which may reduce the cost per unit of these materials. They may also be able to spread fixed costs (like the cost of factories or expensive machinery) over more units, reducing the fixed cost per panel.
- Better Quality Control: With more experience in production, manufacturers are likely to improve their quality control processes. This can reduce the rate of defects and the costs associated with them (like rework or waste), leading to more efficient production.
- Supply Chain Optimization: As the solar panel industry grows, there could be increased specialization and efficiency in the supply chain. For example, suppliers of raw materials or components may find ways to produce their products more efficiently, or logistical processes (like transportation and warehousing) may become more streamlined.
- Workforce Skill Improvements: As employees gain more experience in manufacturing solar panels, they may become more skilled and thus more productive. This can increase the number of panels produced per employee, effectively reducing the labor cost per panel.
- Policy and Regulation Evolution: As the solar industry grows and matures, there may be changes in policy or regulation that reduce costs or barriers for solar panel manufacturers. For example, governments might introduce incentives or subsidies for solar power, or they might streamline the approval processes for new solar installations. These changes could reduce the costs associated with producing or installing solar panels, further improving the learning curve.
In summary, applying the learning curve concept to solar panels suggests that as more panels are produced, various factors may lead to improved efficiency and reduced costs. This is a crucial aspect to consider, especially in the context of renewable energy adoption, as it can lead to more affordable and widespread use of solar power.🔮 Copy this prompt: Apply the concept of Y to X. Give me 7 options. Let’s take this step by step to make sure we have the correct answer.
I would suggest experimenting with the following:
Apply the concept of networks to artificial intelligence. Give me 7 options. Let’s take this step by step to make sure we have the correct answer.
Apply the concept of modularity to innovation in synthetic biology. Give me 7 options. Let’s take this step by step to make sure we have the correct answer.
Apply the concept of technological revolution by Carlota Perez to electricity. Give me 7 options. Let’s take this step by step to make sure we have the correct answer.
Prompt 3: Explore scenarios
We’re not able to predict the future; ChatGPT even less so. But it can help us formulate scenarios and know what to look out for along the way. In addition, we imitate the CoT reasoning by asking AI for its assumptions. This way, we have (1) a higher quality of answers by having it dissect its own output, and (2) an easier way for us to evaluate the answers.
CS: Let's analyse the future global adoption of electric vehicles by 2050. Give me multiple scenarios, each with a precise percentage of adoption and the assumptions underlying the prediction.ChatGPT: Sure, here are three different scenarios for the global adoption of electric vehicles by 2050. These are hypothetical scenarios based on a variety of potential developments and influences. As an AI model, I'm making educated estimations based on current trends and future possibilities. The actual future could vary significantly depending on countless variables.
Scenario 1: Slow Adoption - 40% Global Electric Vehicle Market Share by 2050
In this scenario, the transition towards electric vehicles (EVs) faces several significant barriers, including:
- Infrastructure development: EV charging infrastructure grows, but at a slower pace than needed due to various regulatory hurdles, insufficiant investment, and other issues.
- Technological innovation: Battery technology improves, but still doesn't match the energy density, charge speed, and cost of traditional fossil fuels.
- Consumer behavior: Many consumers remain skeptical about the reliability, utility, and cost of EVs, continuing to prefer internal combustion engine (ICE) vehicles.
- Policy: Global policies do not strongly favor EVs, with only moderate incentives or regulations in place to promote their adoption.
Scenario 2: Moderate Adoption - 70% Global Electric Vehicle Market Share by 2050
In this middle-ground scenario, EVs become mainstream but don't completely replace ICE vehicles:
- Infrastructure development: EV charging infrastructure significantly expands, although certain regions lag behind.
- Technological innovation: Batteries improve in terms of cost, energy density, and charge speed, making EVs a more feasible option for most consumers.
- Consumer behavior: A significant portion of consumers embrace EVs due to their environmental benefits, performance, and reducing costs, but a sizeable group remains attached to ICE vehicles.
- Policy: Policies worldwide increasingly favor EVs, with subsidies and tax incentives for buyers, plus emission standards that make ICE vehicles less attractive.
Scenario 3: Rapid Adoption - 90% Global Electric Vehicle Market Share by 2050
In this optimistic scenario, EVs become the dominant form of transportation globally:
- Infrastructure development: Fast, universal roll-out of EV charging infrastructure globally, including in developing countries.
- Technological innovation: Significant breakthroughs in battery technology occur, leading to EVs that are cheaper, more energy-efficient, and quicker to charge than their ICE counterparts. Advancements in renewable energy also make EVs cleaner overall.
- Consumer behavior: Consumers globally opt for EVs due to their cost-effectiveness, efficiency, performance, and environmental benefits. ICE vehicles become increasingly seen as outdated and environmentally harmful.
- Policy: Strong, global policy measures actively promote EVs and disincentivize ICE vehicles. This includes large-scale subsidies for EVs, strict emissions standards, and even bans on new ICE vehicle sales in many regions.
Each of these scenarios comes with its own set of uncertainties, and the actual adoption rate could fall anywhere within this range or even outside of it, depending on numerous factors that are hard to predict accurately.🔮 Copy this prompt: Let’s analyse X by Y. Give me multiple scenarios, each with a quantitative assessment of Z and the assumptions underlying the prediction.
Prompt 4: Question assumptions
When looking at the future, a key challenge lies in questioning our preconceptions and resisting the temptation to only search for information that validates our initial ideas. In the previous example, we looked at scenarios and asked for the assumptions underlying each one.
Now, let’s push the boundaries further. Suppose a pessimistic scenario, something you consider as improbable, and inquire from ChatGPT what would need to be true for it to occur.
ChatGPT often gives broad responses to such inquiries, hence it proves beneficial to incorporate descriptive words like “tangible”, and ask it to provide measurable aspects.
CS: What would need to be true for AI not to enable economic growth? Give me a list of tangible elements, and how to measure them. ChatGPT: To provide a list of tangible elements that would need to be true for AI not to enable economic growth, along with potential measurement approaches, we can consider the following factors:
- Limited Adoption Rates: Measure the rate at which businesses and industries are adopting AI technologies. Low adoption rates indicate a lack of integration and utilization of AI in various sectors.
- Stagnant Productivity: Monitor productivity growth across different industries and compare it to historical trends. If productivity remains stagnant or declines despite the availability of AI technologies, it suggests that AI is not contributing to economic growth.
- Lack of Investment: Analyze investment patterns in AI research and development, infrastructure, and implementation. Reduced investments in AI-related initiatives and projects indicate a lack of confidence in its potential for economic growth.
- Skill Gap and Workforce Disruption: Evaluate the availability and quality of AI-related skill sets in the labor market. Measure the number of professionals trained in AI, data science, machine learning, and related fields. If there is a significant shortage of skilled professionals and widespread job displacement without adequate retraining programs, it suggests AI is not driving economic growth.
- Limited Technological Advancements: Assess the rate of AI-related breakthroughs and advancements. If there is a lack of significant progress in AI algorithms, hardware capabilities, or other related technologies, it indicates a potential hindrance to economic growth.
- Negative Impacts on Employment: Track employment rates and job creation in industries where AI is expected to have a significant impact. If AI adoption leads to a decline in employment opportunities without the emergence of new job sectors, it suggests AI's limited contribution to economic growth.
- Regulatory Constraints: Evaluate the presence of restrictive regulations and policies that impede the development and deployment of AI technologies. Assess the level of compliance burden on businesses and the restrictions on AI innovation. If regulatory barriers are excessively high, it can hinder economic growth.
- Negative Externalities: Identify any negative consequences associated with AI adoption, such as ethical concerns, privacy issues, bias in decision-making, or social inequality. Measure public sentiment, conduct surveys, and monitor media coverage to gauge the perception and impact of these negative externalities.
By monitoring and assessing these tangible elements, policymakers, economists, and researchers can gain insights into the extent to which AI is contributing to or inhibiting economic growth. These measurements can help inform decisions and actions to address any potential barriers and maximize the positive impacts of AI on the economy.🔮 Copy this prompt: What would need to be true for [unlikely event] to happen? Give me a list of tangible elements, and how to measure them.
Two bonus tips + invitation:
Ask for relevant historical examples when ChatGPT gives you any prediction or difficult-to-substantiate claims. If it can not supply you with multiple, high-quality historical examples of similar situations, it may be worth questioning the premise.
Ask ChatGPT to design a prompt for you.
You may ask ChatGPT to help you design a prompt, according to your specific needs. Ethan Mollick has usefully designed this “mother” prompt:
EM: Help me craft a really good prompt for ChatGPT. First, ask me what I want to do. Pause and wait for my answer. Ask questions to clarify as needed. Second, once you have the information suggest a prompt that include context, examples, and chain of thought prompting where the prompt goes step by step through the problem. Third, show what your response as ChatGPT would be to the prompt. Fourth, ask if the user has any suggestions and help them revise the prompt.
ChatGPT: Of course, I’m here to assist you! To start off, could you please tell me what you would like to accomplish with the prompt for ChatGPT? Are there any specific topics, themes, or tasks you have in mind?
Join a member-only AI in Practice online event on 27 July. Members will showcase how they use AI in their work and life. RSVP here — the event is open only to members with an annual subscription.
I would love to hear and learn from you, dear members.
What lessons have you learnt from your experiments?
What are your favourite prompts?
Great idea, I would love to add some insights from my experience ( Behavioral Mathematican, Lecturer in data sience, AI Expert and a prompt Nerd). A basic prompt of 2 or 5 lines, does not give you the insights which are possible. It's a starting point but never the full potential. So I would use something like this for GPT4 ( yes the model matters for the prompt and the approach) :
In this rapidly evolving, volatile, uncertain, complex, and ambiguous (VuCA) world, it can be challenging to identify and understand the potential interconnections. Given these circumstances, I would like to explore the potential relationship between X and Y. Please bear in mind that the absence of a relationship is a valid conclusion and the goal here is to avoid confirmation bias or other cognitive biases while ensuring a fact-based investigation.
If a plausible relationship between X and Y exists, please illustrate with up to 7 fact-based examples. Each example should clearly demonstrate the connection, contain accurate and detailed information, and where possible, highlight the source of the information.
Furthermore, if possible, please provide insight into whether these relationships are causative, correlative, or possibly coincidental, based on the information available.
It would be greatly appreciated if you could also discuss the potential implications and broader context of these relationships within our interconnected world, in order to encourage a comprehensive understanding
Happy prompting
Thanks for this Promptpack. Useful and informative. My favorite from two months ago is temporarily misplaced. The important thing about it is that it failed to resolve my concerns about material in David Abulafia's "The Endless Sea, a Human History of the Oceans." The facts in question concerned the value of the saffron trade between San Gimignano and Alepo. I will look into it more soon, but my current impresion is that critical facts are 1) in Italian, Greek, Arabic, Farsi and 2) not in web pages. Books and papers are less likely to be part of GPTs training than web pages in English. Therefore the type of facilities offered by Langchain may be valuable. I can report that tonight, ChatGPT gave better results than in the past, but it seems stuck on the idea that spices went from East to West. Abulafia's book says the quality from San Gimignano was so high, that it was valued even in Perisa. I'll have time soon to look at more of you suggested prompts. If San Gimignano's towers were financed in large part by selling saffron to Aleppo and then to Persia, the world should know. Seems like taking coal to Newcastle.