LLM Note #1

I rely heavily on LLMs in my production workflow process. A notable challenge is the in-built sycophancy. I find there are a couple of available remediation options:

1. Specifically ask for a hostile structural review.

2. Tell the LLM that the idea is external to you. For added effect, tell it that you are sceptical of claims being made.

By signalling that externalising the idea couched in scepticism will refocus its sycophancy and force it to side with your incredulity. This will likely result in an ‘aren’t you clever to spot this weakness no one else might have‘.

Try it. It works, but don’t blame me if it ruins your relationship with your AI companion.