Impromptu
Subtitled 'Amplifying Our Humanity Through AI'; co-written with GPT-4
Summary
Reid Hoffman wrote this book with GPT-4 during the months before the model's public release, and the format is the argument. Each chapter presents Hoffman's prompts and the model's unedited responses, with his commentary on what the output reveals, including where it is wrong. The chapters cover education, journalism, creative work, law, medicine, social media, and the work of building and using such systems.
Hoffman's position is that large language models will function as amplifiers of individual capability and that the appropriate response is broad early deployment rather than restriction. He argues that iterative public release produces better safety outcomes than closed development, because problems surface where they can be addressed. He is explicit that he holds financial interests in the companies concerned, including a board seat at OpenAI at the time of writing, and readers should weigh that throughout.
The book is a period document. The specific model behaviour it demonstrates, including the hallucinations Hoffman deliberately preserves, describes a system that has been superseded several times over. Its arguments about labour displacement are optimistic and lightly evidenced, and it does not engage seriously with the strongest safety objections. Read it as a record of how a well-placed investor understood the technology in early 2023, not as an analysis of where the technology went.
Key takeaways 7
- Hoffman argues that language models amplify individual capability rather than replacing categories of worker.
- Iterative public deployment surfaces failure modes that closed testing does not, so releasing early is the safer policy.
- The models produce confident false statements, and Hoffman prints examples rather than editing them out.
- Professional work in law and medicine will shift toward supervising model output and away from producing first drafts.
- Access to these tools, rather than the tools themselves, will determine how their benefits are distributed.
- Hoffman treats regulatory restriction of model development as more likely to entrench incumbents than to improve safety.
- Human judgement about which questions to ask becomes the scarce input once generating answers becomes cheap.
Who should read it
Investors and founders who want a documented snapshot of how the technology and its advocates presented themselves at the start of the current cycle.
About Reid Hoffman
Co-founder and former executive chairman of LinkedIn; partner at Greylock Partners; co-founder of Inflection AI
Blitzscaling and The Start-up of You outlived their news cycles and gave the industry durable vocabulary, on top of a first-rate operating and investing record.
Similar books 6
Chosen by shared topics, the entries that hold them, and the people who recommended both.
Recommended by: Rana el Kaliouby
Why China may match the US in AI now that applying it matters more than research.
Recommended by: Balaji S. Srinivasan, Alfred Lin and 2 others
AI research from Dartmouth in 1956 to the funding cuts, told by the field's leaders.
The founder scene around OpenAI, reported from inside the funding rounds.
Recommended by: Patrick Collison
Where it sits in the reference 2
Topics