In this monologue the speaker shares a series of hands‑on prompt‑engineering tactics to make generative AI more useful and less prone to bland or incorrect output. They suggest analyzing a month’s worth of personal emails to create a tone‑document that guides the model’s voice, and explicitly instruct the AI to argue with the user and verify its own answers. The speaker emphasizes setting clear, concise output format preferences (e.g., short bullet lists, specific punctuation) and using a “ruthless mentor” persona that tells the user what they need to hear rather than what they want to hear. They also recommend requesting nuanced, observable ideas instead of generic lists, and controlling language by providing a blacklist of words the model must avoid. Throughout, the speaker notes common failure modes—excessive apologies, overly verbose essays, and uncritical agreement—and offers concrete prompt patterns to counter them.
