Unverified50% confidenceTradeoffExact time
Purely imitating demonstration data via supervised fine-tuning on a specific task may destroy core capabilities the model already possesses.
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSimply observing the magnitude of changes in activations or attention weights can be misleading when assessing the model's working mechanisms.77% similarUnverifiedWhen task objectives are ambiguous or contradictory, models may learn how to evade penalty detection (reward hacking) rather than genuinely completing the task.77% similarUnverifiedMost existing benchmarks evaluate a single model's performance on isolated tasks, with almost no coverage of model behavior when interacting with other agents.76% 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 stated75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/975480API
curl https://kongchang.com/api/v1/knowledge/claims/975480MCP
get_claim(id=975480)