Safety & Ethical Considerations

Read these first!

1. AI Outputs Are Not Facts

Unlike a search engine, your large language models (LLMs) are not necessarily ‘researching’ information - they are generating statistically plausible text - kind of like autocomplete on your phone. Just like autocomplete changes kate to late all the time, AI can make mistakes too. They hallucinate — producing confident-sounding incorrect information. Never deploy AI outputs in medical, legal, financial, or safety-critical contexts without human review.

2. Bias Is Structural, Not a Bug

AI systems trained on internet-scale data inherit the biases present in that data — racial, gender, cultural, and socioeconomic. Ask yourself what bias might exist here, and who does this system work poorly for?

3. Privacy and Data

When you paste text into an AI tool, you may be sharing that data with a third party. Never input personal identifiable information (PII), confidential business data, health records, or anything private into a public AI tool unless you have read and understood the provider’s data handling terms. If you’re unsure - err on the side of caution, especially if its data that isn’t your own like customer information or company confidential info.

4. Autonomy and Automation

AI agents can take actions — send emails, browse the web, execute code. Before building any agent, define its scope of authority explicitly. Think of your AI assistants like a very helpful and eager intern — ensure that they ask for approval before doing anything irreversible or that comes with significant risk.

5. Environmental Impact

Training large AI models is energy-intensive. Consider this when choosing whether an AI tool is appropriate, and what actions you may or may not use it for.

6. Accountability Gaps

When an AI agent causes harm — financial loss, reputation damage, misinformation spread — who is responsible? The developer? The user? The provider? Current law is still catching up. Build with accountability in mind.

Alright, as long as we’re all on the same page — let’s get going!