An Opinionated Guide to Using AI (www.oneusefulthing.org)

This piece is an updated, opinionated how-to for late‑2025 that leverages OpenAI’s usage data (about 10% of people use AI weekly and much of that is information-seeking) to give practical guidance on which AIs to pick and when to upgrade. The short advice: for casual info and high-volume free use pick your favorite free front end (many are powered by a small set of core models). For a low‑cost serious tier ($20/month) choose among OpenAI’s ChatGPT (GPT‑5 family), Google’s Gemini, or Anthropic’s Claude (Sonnet 4.5); heavy technical users may need $200 tiers or specialized offerings like xAI’s Grok. Open‑weight families (Deepseek, Kimi, Z/Qwen, Mistral) and aggregation services (Copilot, Perplexity) are competitive for free/cheap use. Key distinctions: model variants (chat vs agent vs “wizard”), access to web search, privacy/training opt‑outs (ChatGPT and Claude allow blocking training; Gemini is more limited), and differences in personality and safety posture. For real work, use agentic models (they autonomously run multi‑step tasks) or Deep Research modes (10–15 minute, citation‑heavy web research) and connect the model to your data (Gmail, drives, calendars) for briefings and higher accuracy. Multimodal features—voice, image/video upload, screen sharing—are mainstream: Gemini leads image generation and has Veo 3.1 video, OpenAI has Sora 2 (social-media style video), and Claude excels at document/Office outputs. Caution: hallucinations are reduced but persistent; media is increasingly untrustworthy as video/image generation becomes indistinguishable from real footage.