ARTropic’s latest flagships may not be used to train

The latest, Claude 3.7 sonnet, Claude’s Claude, costs of a million dollars “to train less than 10 ^ 26 computation power. It is based on Professor WHATONTON Ethan Molick, entering the X on Monday that is prepared clarification accepted from PR. “I have been contacted by an anticropic that tells you that sonnet 3.7 will not be considered 10. ^ Claimer flop and cost some tens of millions of dollars,” she wrote, “although the model will larger.” The apwart is approaching anthropic to confirm but has not received a response as a publication period. Assuming Claude 3.7 sonnet is just “some tens of millions of dollars” to practice, not focused on the relevant fee, it is a sign of how low it will be a model model. Claude 3.5, before the sonnet, released in the fall of 2024, equal to several million dollars to train, the AMOdei’s Dairio Amodei in the new essay. The number of these compared to the same price for the 2023 training price training price. To develop the GPT-4 model, the openai spend more than $ 100 million, according to the Openai Sam Altman. Meanwhile, Google is spending up to $ 200 million to practice Ultra Gemedi Models, Stanford Learn. They say, amodei expect ai ai model for billion dollar charges. Of course, the training cost does not capture work like a basic safety and basic research. In addition, because the AI ​​industry gathering “the model” the “reason” can be done with a number of problems for the time, the model computational compartment can continue to increase.

Source link