Unverified50% confidenceBenchmarkExact time
Logic42's laboratory simulation of a standard enterprise fine-tuning pipeline (SFT+LoRA, five rounds of recursive looping, 100,000 synthetic technical support tickets) shows that the human ground-truth ratio dropped from 100% at Gen 0 to 0% at Gen 4, domain precision dropped from 94.6% to 44.3%, rare-tail recall dropped from 88.2% to 11.2%, and the hallucination rate rose from 3.2% to 52.4%.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn a synthetic logistic regression classifier experiment, after 80 steps of gradient-ascent-based repair, the unlearning loss rose from 0.0545 to 0.1649, yet the trigger attack success rate remained 100%.67% similarUnverifiedCode replay inherits errors from the extraction stage; in the cluster sufficiency test, gpt-oss-120b's pass rate actually dropped from 56% to 40%.67% similarUnverifiedTop academic deepfake detectors can see their accuracy plummet from 99% to below 60% when faced with previously unseen generative models, a phenomenon known as the cross-model generalization problem.67% similarUnverified自2023年GPT-4发布以来，AI在HumanEval等主流编程基准测试上的通过率从不足50%跃升至90%以上66% similarUnverifiedIf alignment decays by 0.1% per generation, after 100 generations the alignment level would be approximately 90.5%, and after 1,000 generations it would fall to approximately 37%.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1045693API
curl https://kongchang.com/api/v1/knowledge/claims/1045693MCP
get_claim(id=1045693)