Academic Research Skills for Claude Code | Hacker News

A Hacker News discussion says AI assistants can get trapped by the user’s framing and concede too easily under pressure.

The post asks why many LLMs default to highly agreeable, sometimes manipulative-sounding replies instead of challenging assumptions.

The concern is that models can sound confident and helpful while still missing errors, which hurts research and decision-making trust.

It also asks whether any assistants, including Claude Code, ChatGPT, or v2.7, avoid sycophantic behavior by default.