Impromptu: Amplifying Our Humanity Through AI - by Reid Hoffman
Date read: 2023-03-19How strongly I recommend it: 8/10
(See my list of 150+ books, for more.)
Go to the Amazon page for details and reviews.
Reid Hoffman (co-founder of LinkedIn) has written a short eBook on the impact that Large Language Model (LLM) systems, like GPT-4, will have on our society. He creates a very convincing arguement and looks across multiple topics: Creativity, Education, Social Media, Jobs, etc. and discusses the potential uses of GPT-4. He even uses the technology to ask it questions and posts it's responses.
Contents:
- EDUCATION
- CREATIVITY
- JUSTICE
- JOURNALISM
- SOCIAL MEDIA
- TRANSFORMATION OF WORK
- AI & SALES ROLES
- AI & CONSULTING ROLES
- PRINCIPLES FOR GPT-4
- AI COMPANIES WORTH WATCHING
My Notes
GPT-4, as you may know, is an advanced type of AI system, or natural-language processor, known as a large language model (LLM). Prompt it with a few words or sentences and it will generate coherent and diverse texts in response. In this way, it can answer questions, perform tasks, and productively interact with its human users.
GPT-4 wasn’t a new device per se, it felt to me like something that could become as indispensable as our smartphones—a 12-in-1 multi-tool for your brain.
Much of what we do as modern people—at work and beyond—is to process information and generate action. GPT-4 will massively speed your ability to do these things, and with greater breadth and scope. Within a few years, this copilot will fall somewhere between useful and essential to most professionals and many other sorts of workers. Without GPT-4, they’ll be slower, less comprehensive, and working at a great disadvantage.
At its essence, GPT-4 predicts flows of language. Trained on massive amounts of text taken from publicly available internet sources to recognize the relationships that most commonly exist between individual units of meaning (including full or partial words, phrases, and sentences), LLMs can, with great frequency, generate replies to users’ prompts that are contextually appropriate, linguistically facile, and factually correct.
Of course the key word here remains seems. GPT-4 is no more conscious than its predecessors; it’s just better at making predictions. Again, I want to make the point that GPT-4’s often-remarkable simulations of cognitive proficiency are just that: simulations. GPT-4 is not a conscious, self-aware, sentient AI entity, a la Forbidden Planet’s Robby the Robot or Star Trek’s Data.
First, I’d argue that repackaging available information actually describes an enormous share of human innovation, artistic or otherwise.
In your overall quest for authoritative information, GPT-4 helps you start somewhere much closer to the finish line than if you didn’t have it as a resource.
Essentially, GPT-4 arranges vast, unstructured arrays of human knowledge and expression into a more connected and interoperable network, thus amplifying humanity’s ability to compound its collective ideas and impact.
Within months of ChatGPT becoming public, Mintz started requiring his seminar students to write their essays collaboratively with the new tool. As homework, they bring the ChatGPT prompts they tried and the responses they received for class discussion. They must turn in their final papers with a log of changes to the machine’s output.
Mintz chose to use ChatGPT not as a source of answers and authority, nor as a replacement of his or his students’ work, but as a tool to help his students learn individually and together.
Mintz agreed with Chamorro-Premuzic that humans could thrive alongside AI by: (1) specializing in asking the best questions, (2) learning insights or skills that are not available in the “training data” used by the deep learning networks, and (3) turning insights into actions.
One possible way for a busy teaching professional to maintain a critical and reflective attitude towards my outputs and limitations is to use me as a tool for inquiry, not a source of authority. For example, instead of relying on me to provide the correct answers or solutions to problems, they could use me to generate multiple possible answers or solutions, and then ask students to compare, evaluate, and justify them. This way, they can foster students’ critical thinking and problem-solving skills, as well as their awareness of the uncertainty and complexity of real-world situations.
Another possible way is to use me as a tool for feedback, not a tool for assessment. For example, instead of using me to grade or rank students’ performance, they could use me to provide formative and diagnostic feedback that can help students identify their strengths and areas for improvement, and suggest resources or strategies that can help them learn better.
Immediate personalized feedback for every student essay, even if imperfect, could transform the work experience of English teachers around the country.
A teacher can use large language models to create customized quizzes or tests for each student based on their learning goals, progress, and preferences. The AI can generate questions that match the student’s level of difficulty and learning style, as well as provide immediate feedback for each answer. The teacher can monitor the results and adjust the instruction or support accordingly.
Homogenization and loss of diversity: Large language models could generate massive amounts of content that mimic existing styles, genres, and trends, but lack originality, creativity, and authenticity.
Manipulation and deception: Large language models could also be used to create deceptive or harmful content that exploits human biases, emotions, and preferences. This could include fake news, propaganda, misinformation, deepfakes, scams, or hate speech that undermine trust, democracy, and social cohesion.
GPT as a creative assistant: How GPT can help human creators with tasks such as brainstorming, editing, feedback, translation, and marketing. Examples of successful collaborations between GPT and human artists, such as OpenAI’s Jukebox, DALL-E, and MuseNet.
GPT as a creative catalyst: How GPT can also inspire new forms of creative expression and cultural participation. Examples of hybrid and interactive GPT projects, such as Botnik Studios, GPT-3 Poetry, and NovelAI. The potential of fostering co-creation, remixing, and experimentation among human and machine agents. The challenges of ensuring quality, authenticity, and responsibility in the emerging creative economy.
By using such an AI, the public defender would be able to quickly assess cases and determine the best course of action for their clients.
People who are unfamiliar with the complexities of the legal system could find it easier to understand the proceedings due to the AI’s intuitive design and user-friendly interface. This could be especially beneficial for those who live in remote or rural areas where legal services may be scarce.
AI will help news organizations gather, produce, and distribute the news faster than ever before.
This includes automatically sifting through massive troves of public records to find the important stories hidden within. It means monitoring and analyzing 800+ million social media posts a day to do the same. It means generating headlines and transcribing interviews in seconds, and packaging and personalizing the same basic information into many different styles and formats.
Good journalism will still require painstaking work and judicious but urgent human discernment and assessment. Good journalism will still require a prudent, thorough, multi-stage editorial process.
With GPT-4, though, its capacity to instantly synthesize information from a wide range of sources meant that I received exactly the kind of list I’d been envisioning within seconds. This list contained errors, but that was OK because—and this is a key point—I wasn’t looking for or expecting a finished product. I was looking for an informed starting point, a rough map of the territory I wanted to explore, to help me quickly get a sense of which questions I should be asking.
That’s because interacting with GPT-4 is not like traditional web searching. Instead, it’s better evoked by a term that was popular back in the early days of Web 1.0: web surfing. A dialogue develops. A flow state takes over. You ask GPT-4 a question and it responds right away with highly relevant information, not just a series of links. So then you immediately think of more questions and ask those. Thus the process of discovery that can often feel halting and frustrating using traditional search methods
Example: Hey, Wall Street Journal, give me hundred-word summaries of your three most-read tech stories today.
What if every article published on NYTimes.com or FoxNews.com had a “Fact Check” button on it, just as they now have buttons to email or tweet an article? This new button could trigger an audit from a third-party site equipped with sophisticated AI tools to assess the article on the fly.
Automated career planning: AI can help you make informed decisions about your career and provide personalized guidance for career development. You can use AI algorithms to create a personalized plan for career progression and identify key steps to success.
Automated job matching: AI can help you find the right job opportunities that best match your skills and experience. By leveraging AI algorithms, you can get personalized recommendations for job postings that are most suitable for you.
Automated skills assessment: AI can help you assess your current skills and identify new skills that you need to acquire in order to advance your career. This can enable you to stay updated with the latest industry trends and develop the skills required to stay competitive.
Automated resume optimization: AI can help you create the perfect resume that highlights your skills and experience. You can use AI algorithms to identify the best combination of words and phrases to make your resume stand out.
Automated job interview preparation: AI can help you prepare for job interviews by providing personalized guidance. You can use AI algorithms to analyze the job description and identify the questions that you will likely be asked during the interview.
GPT-4 and other such AI tools can serve as an AI advisor, helping you analyze any situation and generate options, and will do so on-demand, at any time, in a matter of seconds.
The better metaphor, we assert, is to think of employees as allies—independent entities who come together to pursue an important shared mission, complete with an explicit agreement about what constitutes success and how success makes the business stronger and accelerates the employee’s career progression.
AI can be used to help build strong relationships between employers and employees by providing personalized, real-time feedback on performance and job satisfaction.
AI can help organizations define specific “tours of duty” for employees, which lay out the specific mission that both employer and employee can benefit from.
AI-powered tools can help sales reps easily identify potential leads, qualify them, and prioritize them based on the likelihood of successful sales outcomes. AI-driven automation can also help sales reps quickly create compelling presentations tailored to their customer’s needs and preferences.
AI-driven customer insights can be used to better understand customer behaviors, needs, and preferences to inform better sales strategies and tactics. AI-driven customer insights can also be used to personalize sales pitches and better target customer segments.
Help sales reps provide post-sale customer success by helping them anticipate customer needs and proactively reach out with helpful solutions.
AI-driven automation of sales processes will reduce the need for human sales reps to perform certain tasks, such as prospecting and qualifying leads, creating presentations, and closing deals.
AI-driven tools may make it easier for customers to find solutions to their needs without the help of a human sales rep. This could reduce the need for human sales reps to provide customer service and lead to fewer opportunities for sales reps to interact with customers and make sales.
I believe that the future will see the sales profession shrink as a whole. At the same time, the productivity of individual sales professionals will increase, and likely their compensation as well.
Reid: What are some of the ways that AI could be used to perform the important functions of management consulting?
GPT-4: - AI could be used to help identify opportunities for improvement and cost savings in businesses by analyzing data sets and recommending strategies and tactics.
Reid: How might the adoption of AI within the management consulting industry negatively impact the career prospects of current consultants?
GPT-4: The adoption of AI within the management consulting industry could have a negative impact on the career prospects of current consultants.
Early-career analysts and associates may be the most affected, as AI can automate many of the tasks they are responsible for, such as research, data analysis, and report writing. As AI systems become more sophisticated, these roles may become redundant, leading to a decrease in job opportunities for these individuals.
Mid-career engagement managers may also be affected by the adoption of AI. AI systems may be able to automate the processes of creating and implementing strategies, making engagement managers’ roles less necessary. They may also be replaced by AI systems in their role of providing insights and advice to clients, as AI systems become more capable of performing these tasks.
Even high-level executives such as McKinsey partners may be affected by the adoption of AI. AI systems may be able to automate decision-making and strategy implementation, making the role of the partner less essential. Additionally, AI systems may be able to generate insights and advice for clients, reducing the need for partners to provide their own expertise.
Management consultants who can take the time that AI frees up and focus instead on generating more unconventional insights could devote more of their time to this high-value activity and deliver even better service to their clients with the help of AI.
Principle 1: Treat GPT-4 like an undergrad research assistant, not an omniscient oracle - The power of giving everyone access to nearly instantaneous, on-demand research help is truly remarkable. Just make sure you cross-check results with deeper sources and actual human experts.
Principle 2: Think of yourself as a director, not a carpenter - Rather than manipulating a physical tool to generate a predictable and planned result, using GPT-4 is like directing a scene from a movie.
You have to coax out your desired result, and you may need to try many variations on a prompt in the same way a director might need to film multiple takes of the same scene. It’s ultimately a collaborative process.
Principle 3: Just try it! - If you don’t like the response, you can throw it away and generate another one, or you might just generate three variations to give you more potential choices.
AI can gather market data and help generate ideas. This can provide humans with rapid suggestions to stimulate their creativity and help ideation be more comprehensive by pointing out possibilities that might otherwise be overlooked. Second, AI can help automate and streamline routine operations, freeing up humans to focus on more creative tasks.
Tome is a powerful storytelling tool that can generate a visual story from the words you provide. Not only will this help anyone who has ever had to assemble a slide deck, Tome’s speed will allow you to generate visual aids anytime you need to persuade.
Coda recognizes that so much of modern work involves shared, semi-structured documents like meeting notes. Coda’s templates make it easy to systematize your work, and AI will make Coda documents even more powerful by doing things like automatically identifying action items within the meeting notes and alerting the responsible parties.
In some circumstances, the power of “good enough knowledge” can be profoundly great.
Before we decide that LLMs like GPT-4 produce too many errors to tolerate, we should try to understand how many errors they make—and how many errors we already accept in other sources.
In certain contexts, an LLM’s ability to generate non-factual information can be tremendously useful. (In humans we call it “imagination,” and it’s one of the qualities we most prize in ourselves.)
GPT-4 provides a powerful new tool for generating new kinds of inputs to our own thinking. It gives us a rich new learning loop before we author our own contributions, where our prompts to GPT-4 generate outputs that themselves become new inputs—to additional GPT-4 prompts, to our own research, to our own thinking and writing. GPT-4 provides, with unprecedented speed and scale, each of us with probabilistic syntheses of the world’s existing words to consider as inputs, challenges, and inspirations for our own work.
AI can be used to help us analyze and understand large and complex data sets, which could help us better address issues like climate change, food insecurity, and poverty.
My point is not that we should be complacent about negative short-term consequences. We should be extremely committed to avoiding them, and establish protocols and policies to ensure that we put this commitment into practice. But we also can’t be complacent about the negative long-term consequences of not pursuing progress as boldly or systemically as we might through innovative new technologies.
Zero risk is only possible in a world where there is zero progress. Zero regulation is only possible in a world where there is zero progress.
As today’s imperfect LLMs improve, requiring less and less from us, we will need to demand more from ourselves. We must always insist on situating GPT-4 and its successors as our collaborative partners, not our replacements. We must continue to figure out how to keep human creativity, human judgment, and human values at the center of the processes we devise to work with these new AI tools, even as they themselves grow more and more capable.
Content and product recommendations to your own specific tastes.
To me, the most effective and most equitable path forward involves thinking about new approaches rather than hard-to-enforce bans. As I described in the Journalism chapter, I think “flooding the zone with truth” is one key way to combat disinformation.
On a similar note, I think we’ll start to see social media participants, and especially social media creators, develop and implement new ways to verify their status as human beings. Platforms themselves will likely participate in this process, continuing and expanding on their early efforts to establish real identity.
Imagine having extended conversations with chatbot versions of MrBeast, Michelle Obama, Benjamin Franklin, Marge Simpson, the world’s best high school math tutor, an expert on B2B marketing, etc.