arXiv cs.AI· Sujato Dutta, Padala Indira Sneha, Kypu Mahitha, Shashank Vanga, Neeraj Choudhary·· 3 小时前AI 评分39
Third Eye：递归自我改进不能只看即时收益
Third Eye: Beyond Immediate Gains in Recursive Self-Improvement
AI 导读
论文提出 Third Eye 方法，把模型的即时性能与继续二次原生更新后的结果分开评估。在 Qwen3-4B、Llama-3.2-3B 等 3-8B 模型的数学与代码实验（64 次全新运行）中，即时选出的效用接近零，而继续更新使其再降 1.78 个百分点（95% 区间 [0.68, 2.88]）。结论支持用独立评分和无更新基线、以适应转变视角评估持续性自我更新。
正文 · 原文
Abstract:Does a useful self-update remain useful after the model updates again? Third Eye separates immediate performance from the outcome of a further native update. We evaluate five 3-8B backbones on math and code: 64 fresh runs, three candidates per run, and three continuations per candidate. Selection and final scoring use disjoint items. In the balanced Qwen3-4B/Llama-3.2-3B comparison, immediate-selected utility is near zero, but continuation reduces it by 1.78 percentage points (simultaneous 95% interval [0.68, 2.88]). Candidate-averaged immediate utility is already negative: selection partly accounts for the apparently neutral first step. Continuation also lowers utility for SmolLM3-3B and Llama-3.1-8B; Qwen3-8B remains inconclusive. Target and out-of-distribution losses dominate retention changes. Matched recipe and recipient-donor controls identify neither a reliable mitigation nor a single mechanism. The results support evaluating persistent self-updates as adaptation transitions, with independent scoring and a no-update baseline, rather than interpreting immediate performance as their full value.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.12988 [cs.LG] |
| (or arXiv:2611.12988v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.12988 arXiv-issued DOI via DataCite |
Submission history
From: Sujato Dutta [view email]
[v1]
Sun, 8 Nov 2026 10:14:16 UTC (159 KB)
来源：arXiv cs.AI · arxiv.org