A look at some early-adopters using ChatGPT, GPT-3, and other text generator bots to write business emails, find creative inspiration, and more
The latest AI sensation, ChatGPT, is easy to talk to, bad at math and often deceptively, confidently wrong. Some people are finding real-world value in it, anyway.
Context & Ripple Effects
This piece lands at the very start of the generative-AI adoption curve: weeks after launch, [[a:1157423|early adopters are already routing business emails and creative brainstorming through ChatGPT and GPT-3]], treating a chatbot as a drafting tool rather than a demo. The catch the article flags — fluent output that is 'deceptively, confidently wrong' — is the same confabulation problem experts later traced to models filling gaps in training data with plausible-sounding words (Ars Technica's confabulation explainer).
What makes the moment worth watching is that the informal habits described here hardened fast: within two months, CEOs and engineers across many companies were formally experimenting with the tools (WSJ on enterprise-wide experimentation), and by mid-2023 a study measured writers completing press releases and reports 40% faster with quality scored 18% higher (MIT Technology Review's productivity study) — evidence the early-adopter anecdotes pointed at something real.
First-order effects
- Individual knowledge workers gain a working draft machine for emails, class material, and creative prompts right now — with the burden of catching confidently wrong output falling entirely on them.
Second-order effects
- Employers move from anecdote to policy as staff experiment at scale, forcing companies to decide which tasks tolerate generative error and which don't — the tolerance question Benedict Evans flagged when calling these systems a step change in AI use cases (Evans on the step change).
Third-order effects
- If the measured productivity gains hold while confabulation persists, work software reorganizes around AI-assisted drafting with human verification as the default QA layer — though OpenAI's own GPT Store data, where subscriber-made GPTs drew just 1.5% of desktop visits (Similarweb via FT), suggests demand concentrates in plain chat rather than bespoke builds.
The trend: Generative text is moving from novelty toy to embedded workplace tool, with adoption outrunning the reliability guarantees needed to trust its output.