37 ms·
The article says that this ability was discovered around opus 4.5, but open-source models like deepseek and glm have already surpassed the capabilities of opus4
by taosu_la 3mo ago
The article says that this ability was discovered around opus 4.5, but open-source models like deepseek and glm have already surpassed the capabilities of opus4.5. But why hasn't this been discovered before? Is it because the scale of computing power is not large enough?
- cubefox 3mo agoMore likely it was because Anthropic invests a lot in interpretability research, unlike other AI companies.