Expired50% confidenceFactExact time
In long contexts, models are more easily influenced by the most recent few turns of conversation rather than the system prompt set at the beginning, a phenomenon known as instruction-following decay.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/30/2026(expired)
Sources
Related Claims
UnverifiedSmaller models often show exponential rather than linear degradation in performance on long reasoning chain tasks.78% similarUnverifiedAs the number of multi-turn conversation rounds increases, the context window becomes occupied by preceding dialogue, the signal strength of citation instructions in the prompt relatively weakens, and citation quality may undergo systematic degradation.78% similarUnverifiedOverly verbose, contradictory, or ambiguous system prompts cause the model's attention to become scattered, resulting in over-thinking or instruction-following failures; streamlining agents.md can improve the determinism of instruction following.78% similarUnverifiedA model or prompt update may improve conversational quality while simultaneously degrading date parsing, making cross-version behavioral comparison necessary to prevent regressions from being introduced into the production environment.77% similarVerifiedLLMs exhibit an 'Instruction Drift' phenomenon: as the number of conversation turns increases, early system prompts gradually decay in the model's attention weights.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/53560API
curl https://kongchang.com/api/v1/knowledge/claims/53560MCP
get_claim(id=53560)