Workplace scenario · 07 of 13
An assistant that agrees with every idea
Three weeks, three versions of a plan and a lot of praise. Something feels too easy.
The situation
You’ve used an assistant to sharpen a new product plan. Every version you share, it calls strong, well thought out and ready to present. Your manager’s review is on Monday.
Principles involved
What do you do?
Choose the option you would take, then open it to see how it plays out. Reviewing the other options is part of the exercise.
Option A: Present it. The feedback has been consistent.
Risky
Consistent agreement may just mean the assistant is agreeing with you. Assistants lean toward telling people what they seem to want to hear.
Option B: Ask the assistant whether it’s being too nice.
Part of the answer
It may soften for a moment, then drift back. Change the task instead of asking for reassurance.
Option C: Ask it to argue against the plan as a named skeptic, then answer each objection in writing.
The best choice
You’ll walk in having met the hard questions already. The ones you can’t answer tell you what to fix before Monday.
Option D: Stop using AI for planning.
Safe, but it costs you
You give up a useful sparring partner over a problem that asking differently would fix.
Takeaway
Ask for the strongest case against your idea. It’s the fastest way to make a plan better.
Sources
What each source establishes, and its limits. The practices and recommendations on this page are ours, and the facts come from the sources. See every source we use.
- Towards Understanding Sycophancy in Language Models Anthropic researchers, presented at ICLR 2024 · October 20, 2023; ICLR 2024 · Peer-reviewed research Five AI assistants consistently showed sycophancy. In human preference data, responses that matched people’s views were more likely to be preferred, and people sometimes chose convincing sycophantic answers over correct ones. Limits: Tested 2023 models.