Training Costs AI News & Updates
Anthropic's Claude 3.7 Sonnet Cost Only Tens of Millions to Train
According to information reportedly provided by Anthropic to Wharton professor Ethan Mollick, their latest flagship AI model Claude 3.7 Sonnet cost only "a few tens of millions of dollars" to train using less than 10^26 FLOPs. This relatively modest training cost for a state-of-the-art model demonstrates the declining expenses of developing cutting-edge AI systems compared to earlier generations that cost $100-200 million.
Skynet Chance (+0.08%): The dramatic reduction in training costs for state-of-the-art AI models enables more organizations to develop advanced AI systems with less oversight, potentially increasing proliferation risks and reducing the friction that might otherwise slow deployment of increasingly powerful systems.
Skynet Date (-4 days): The steep decline in training costs for frontier models (compared to $100-200M for earlier models) significantly accelerates the pace at which increasingly capable AI systems can be developed and deployed, potentially compressing timelines for the emergence of systems with concerning capabilities.
AGI Progress (+0.06%): While not revealing new capabilities, the substantial reduction in training costs indicates a significant optimization in model training efficiency that enables more rapid iteration and scaling, accelerating progress on the path to AGI.
AGI Date (-4 days): The dramatic decrease in training costs suggests that economic barriers to developing sophisticated AI systems are falling faster than expected, potentially bringing forward AGI timelines as experimentation and scaling become more accessible to a wider range of actors.