Unverified50% confidenceSolutionExact time
The playbook recommends removing emphatic phrases such as 'think hard' from prompts, because as reasoning capabilities improve these prompts may actually interfere with the model's default behavior.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/4/2026
First Seen
Valid until: 1/2/2027
Sources
Related Entities
Related Claims
UnverifiedFor reasoning (thinking) models, it is recommended to set the thinking intensity to low or turn it off to avoid consuming thousands of tokens on casual conversation.71% similarUnverifiedDescribing unwanted things in a suggestive manner pollutes the model's attention and instead injects those unwanted concepts into it.71% similarUnverifiedExplicit 'fake bad' instructions produced significantly greater distortion than mere contextual framing cues.71% similarUnverifiedRelying on prompt constraints to prevent dangerous operations is inherently unreliable, because the model may be jailbroken, make misjudgments, or forget constraints during long conversations.70% similarUnverifiedThe author recommends disabling thinking, on the grounds that most use cases for building agents or simple applications do not require deep reasoning, and LangChain's compatibility with thinking mode is not yet complete.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/971119API
curl https://kongchang.com/api/v1/knowledge/claims/971119MCP
get_claim(id=971119)