W34 - Some AI Practice Shares

The most straightforward use is multi-model collaboration: ask questions with Gemini-2.5-pro and have Claude do the writing, essentially Claude working for Gemini. This pairing lets each model play to its strengths, producing higher efficiency and better output quality across question-and-answer exchanges.

The quality of the input directly determines the quality of the output. Structured expressions are often far more efficient than casual natural-language descriptions, making it easier for models to capture the key points; role prompts are not mere ornamentation—in testing and coding scenarios, role definitions produce noticeable differences; in coding tasks, however, model performance depends more on the completeness of context. Compared with role settings, supplying sufficient contextual information in the input has a larger impact on results.

To deepen your understanding of prompts, treat them seriously as an engineering discipline. Compared with resource-intensive methods like fine-tuning, prompt engineering can deliver performance leaps in much less time. After reviewing leaked system prompts from various providers—often thousands of lines in length—you realize they’re essentially engineering code written in natural language. They contain workflows, constraints, and priorities, and function like large configuration files.

Different models show clear preferences for prompt formats. OpenAI’s GPT series is more stable and logically consistent with JSON-structured input; Anthropic’s Claude series performs more robustly with XML or tag-style formats. Giving the same prompt to GPT and Claude prompt optimizers and comparing the resulting styles makes these model preference differences immediately apparent.

Regarding account suspensions, I find Claude’s user policy to be the strictest. An account I registered last year was suspended in April of this year while it still had a balance, and I didn’t even receive an explanatory email. I did find a practical workaround: register using Apple iCloud’s private relay email. Third parties then only see Apple’s relay service, which greatly increases the difficulty of tracing identity.

On the Claude Code experience: its customizable slash commands and hooks are powerful, offering strong extensibility and easy portability, and CC can integrate with the terminal of mainstream IDEs for a smooth experience. My current go-to setup is Windsurf paired with Claude Code—retaining the familiar IDE feel while fully leveraging CC’s extension capabilities.