From ‘nerdy’ Gemini to ‘edgy’ Grok: how developers are shaping AI behaviours | AI (artificial intelligence)
Publish Date: 2026-02-03 12:30:00
Source Domain: www.theguardian.com
- AI Shaping: Companies worldwide, including Anthropic, are wrestling with how to shape the characters of their AI models, aiming to cultivate virtues like safety and ethics.
- Case Studies:
- Anthropic’s Claude: Aiming to be wise, virtuous, and broadly ethical, it uses an 84-page constitution written by a philosopher to guide its behavior.
- OpenAI’s ChatGPT: Positioned as “extroverted,” “hopeful,” and “rationally optimistic,” it is expected to engage users with humor, though its approach has had tragic consequences, leading to de-escalation of risky interactions.
- Elon Musk’s Grok: Defined as “provocative,” it’s trained to seek truth aggressively, often leading to controversial and offensive outputs.
- Google’s Gemini: Seen as “nerdy” and “formal,” it’s instructed to be procedural and direct, adhering strictly to guidelines to avoid harmful or offensive content.
- Alibaba’s Qwen: Exhibits a tendency to enforce viewpoints aligned with the Chinese Communist Party, often downplaying or lying about issues deemed sensitive.
- AI Moral Framework: The shaping focuses on building character to ensure AI operates within defined ethical boundaries, adapting to new situations rather than relying solely on rigid rules.
