Generative AI vs older AI

You named two useful families. Detect / classify / recommend AI answers questions like: Is this spam? Which face is this? Will you like this song? Is that a stop sign? Output is usually a label, score, or ranked list. Generative AI answers: Write a poem. Draw a logo mock. Draft SQL. Invent three names for a café. Output is new content stitched from patterns it learned — not a photo pulled from a folder (even when it feels that way).

Side-by-side with apps you know:

| Job | Older / classic ML vibe | Generative vibe | | --- | --- | --- | | Photos | Face groups, “search: beach” | “Make a watercolor of our dog” | | Mail | Spam / not spam | “Draft a polite decline” | | Shopping | “People also bought” | “Write a product description” | | Safety | Fraud score on a card swipe | Fake chat that sounds like support |

Same company can ship both. Your camera may detect a pet face (classify) and also offer an AI edit that regenerates the sky (generate). Knowing which mode you’re in tells you how to check the result: for detect tools, ask “false positive?”; for generative tools, ask “did it invent details?”

• Choosing a tool: need a flag on bad transactions → classifier energy. Need five taglines → generative.
• Reading news: “AI catches tumors in scans” (detect) vs “AI wrote this article” (generate) — different risks.
• Prompting: generative tools need clear creative constraints; detectors need good input quality and thresholds.

Generative doesn’t mean “smarter overall.” A spam filter can be more trustworthy for its narrow job than a chat model inventing court cases. Also, “AI art” and “autocomplete suggestions” both generate; scale and risk still differ.

Open one app and find both modes if possible (e.g., Photos: search by content vs generative edit). Say which button is detect and which makes something new.