Unverified50% confidenceFactExact time
Anthropic's published blog post indicates that prompting instructions such as "double-check your work" and "be as thorough as possible" backfire on modern models, doubling costs and producing worse answers.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/11/2026
First Seen
Valid until: 12/10/2026
Sources
Related Entities
Related Claims
UnverifiedAnthropic published a blog post on reducing costs and improving performance on the Claude platform, pointing out that instructions habitually used to 'push' the model to work harder are now backfiring.77% similarUnverifiedFull-duplex models are highly sensitive to how instructions are presented, with noticeable performance degradation when behavior must be inferred from a persona rather than explicitly stated69% similarUnverifiedIn an example given by Anthropic, 'Be maximally thorough' caused the model to perform dozens of entirely unnecessary knowledge base retrievals.65% similarUnverifiedIn its best-practices document for Agent design, Anthropic recommends that a tool's error response should include sufficient context to enable the model to self-correct without re-querying the user.64% similarUnverifiedThe Anthropic team found that too many assumptions about model limitations were hardcoded into Agent frameworks, and that these assumptions become outdated after model upgrades, causing a large number of best practices to become invalid.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/895208API
curl https://kongchang.com/api/v1/knowledge/claims/895208MCP
get_claim(id=895208)