Dylan Patel 称 Anthropic 在 Fable 5.2 中有意削弱 RSI 能力
Dylan Patel 爆料 Anthropic 的 Fable 5.2 在芯片设计、生物等领域大幅变强，但被刻意削弱了 RSI 能力，GDPx 分数能看出来。关心前沿安全策略和开源模型差距的人可以看看这条讨论。
SemiAnalysis 的 Dylan Patel 在 X 上表示，Fable 5.2 在芯片设计、软件工程、生物和材料科学领域显著更强，这些类别的评估来自 GDPx 基准，任务由真正的领域专家编写。但该模型在金融、法律、量化交易、RSI 和硬件设计上没有提升。
Patel 指出，Anthropic 明确在 RSI（递归自我改进）相关任务上对 Fable 5.2 做了削弱处理，这一决定直接反映在 GDPx 的分数上。这类有针对性的能力限制通常是出于安全考虑，避免模型具备加速自身改进的能力。
AI 资讯账号 omarsar0 转评认为这是 Anthropic 的有意决策，并表示开源模型需要一款真正擅长 RSI 的模型来施加压力，但他目前测试中尚未找到具备 RSI 火花的开源模型。
To what extent the RSI parts were nurfed? Who knows. It’s an interesting intentional decision from them for sure. Open models are going to eat this space and I am here for it. We just need a really good open model that’s really good at this to start putting some pressure. I am testing but haven’t found a solid open model with sparks of RSI. Dylan Patel @dylan522p Fable 5.2 is significantly better at chip design, software engineering, biology, and material science. GDPx is the only eval that test these categories at this level of detail. True experts make these tasks, unlike other benchmarks. Notably it is not better at finance, legal, quantative trading, RSI, or hardware design. Anthropic explicitly nerfed this model in RSI tasks and it's interesting seeing that reflected in the score. 🔗 View Quoted Tweet
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Anthropic, Fable 5.2, GDPx, RSI.
Source attribution: curated from elvis (@omarsar0).
