What is generative AI?

What it means at work

Most of what people mean by AI at work today is generative AI. You describe what you want, and it produces a draft: an email, a summary, a slide outline, a chart description, a formula. That differs from older software, which only did exactly what it was programmed to do with the data you gave it. Generative systems are flexible, which is their strength, and they can also produce plausible mistakes, which is their weakness.

What the sources say

• Adoption: a Pew Research Center survey found the share of employed adults who say they use ChatGPT for work has risen by 20 percentage points to 28% over two years.
• Built into office apps: Microsoft lists drafting in Word, where Copilot can draft, rewrite, and summarize documents, and in Excel it can analyze data, generate insights, create formulas and visuals.
• Prompts in text: the GPT-3 paper states GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. The paper’s GPT-3 is an autoregressive language model with 175 billion parameters.
• Common uses: Pew Research Center found that, among workers who have used AI chatbots for work, the most common uses are research, editing written content and drafting written content.
• Architecture: Vaswani et al. propose the Transformer, based solely on attention mechanisms.
• The weakness is documented: a survey of the research states that LLMs are prone to hallucination, generating plausible yet nonfactual content.

An example

A project manager types, ‘Turn these rough notes into a one-page status update for leadership, with risks at the bottom.’ The tool writes the page. The manager corrects two facts, tightens the tone and sends it. The AI made the first draft; the person owned the final one.

A common misconception

A common misconception is that generative AI understands your business. It knows only what you show it in the conversation plus general patterns from training. The fix is context: give it the source material, the audience and an example of what good looks like, and check what comes back.

Sources

• Pew Research Center, 34% of U.S. adults have used ChatGPT: “the share of employed adults who say they use ChatGPT for work has risen by 20 percentage points to 28%”
• Microsoft, Microsoft Copilot overview: “Word: Draft, rewrite, and summarize documents”
• Microsoft, Microsoft Copilot overview: “Excel: Analyze data, generate insights, create formulas and visuals”
• arXiv, Brown et al., Language Models are Few-Shot Learners: “GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model”
• arXiv, Brown et al., Language Models are Few-Shot Learners: “we train GPT-3, an autoregressive language model with 175 billion parameters”
• Pew Research Center, U.S. Workers Are More Worried Than Hopeful About Future AI Use in the Workplace: “the most common uses are research … and drafting written content”
• arXiv, Vaswani et al., Attention Is All You Need: “the Transformer, based solely on attention mechanisms”
• arXiv, Huang et al., A Survey on Hallucination in Large Language Models: “LLMs are prone to hallucination, generating plausible yet nonfactual content”