Pillar Two

Smart Choice of AI Tools

For the same task, which service and which tier you use completely changes where your data goes and the risks involved.

Where Does Your Input Go?

This is the most important question to ask before choosing a tool.

In 2023, Samsung engineers pasted confidential source code into ChatGPT, and the data was uploaded directly to OpenAI's servers. The problem wasn't hackers — it was that they didn't realize any content entered into an online AI service leaves their computer. Samsung subsequently banned all external AI use.

For researchers, this risk is especially relevant: unpublished experimental data, research ideas, and collaboration agreement drafts — once submitted, you've lost control over them.

✅

Relatively Safe

ChatGPT Team / Enterprise
Google Workspace (Gemini)

Explicit commitment not to use inputs for model training. Suitable for general everyday work.

⚠️ Commitments cannot be verified; sensitive data should still not be entered.

⚠️

Use with Caution

ChatGPT Free
Gemini Personal

Data may be used for training. Avoid entering any unpublished research content or personal information.

🚫

Recommended to Avoid

DeepSeek, Qwen
Kimi, Ernie Bot

Subject to China's National Intelligence Law, which grants the government authority to access server data. Responses may be censored or contain systematic biases.

Decision Principles

Three questions to help you quickly decide whether a service is appropriate.

1

Is the data sensitive?

Unpublished data, collaboration agreements, personal information — regardless of tier, no online service should receive this. Consider a local model.

2

Which country is this service from?

Chinese services are subject to local law, which grants the government access rights. Even for general content, this risk must be evaluated.

3

Does it have a clear data usage commitment?

At minimum, confirm the provider explicitly states how data is used and whether it's used for training. Free tiers typically offer no such guarantee.

Bottom line: Sensitive data should not be entered into any online service, regardless of tier. Commitments cannot be verified — this is the fundamental limitation of online AI.

Major Model Overview

Each provider has a different focus — choosing the right tool makes all the difference.

Model Developer Best For Risk Level
ChatGPT OpenAI (🇺🇸) Everyday writing, Q&A, code Free ⚠️ / Enterprise ✅
Claude Anthropic (🇺🇸) Long document analysis, academic writing, logical reasoning Free ⚠️ / Enterprise ✅
Gemini Google (🇺🇸) Google Workspace integration, long documents Personal ⚠️ / Workspace ✅
TAIDE NSTC (🇹🇼) Traditional Chinese, Taiwan legal formats, local context Taiwan Official ✅
DeepSeek / Qwen / Kimi / Ernie Various Chinese companies (🇨🇳) — Avoid 🚫

Next

Once you've chosen your tool, learn how to talk to it effectively.

Effective AI Interaction →