I didn't really get to test the last Argon twins, but now a new one is out - Carbon. Today it ranks first in most indices among the public models, including a slight advantage over everyone's favorite Opus. Among the interesting things is the Vending Bench benchmark, in which the model won the maximum points. The benchmark basically requires the model to run a business with vendomats for a simulated year (based on the legendary anthropic experiment, which did it physically). The model has to manage supplies, prices, pay rent, transportation to earn as much as possible. This, by the way, hopefully makes it abundantly clear why the "AI will make me money" mission is meaningless nonsense. The better the AI, the worse it will cope with this task, and the more stupid the user, the more often it will ask for it. Reason: the market and capitalism are the first artificial intelligence in the world, which is already several hundred years old. There are no inefficient markets if no one interferes with them. And no one can stop the spread of artificial intelligence and the markets to become more efficient. Therefore, any opportunity for arbitrage or optimization in the market will be used faster than a few years ago. If a trend goes viral on TikTok, then it takes weeks, even months, for that trend to show up in brand advertising, T-shirt prints, and on Mackenzie. It will be minutes. Any "easy" way to make money (even the smallest, let's say a few dollars that a person won't get out of his chair) will immediately be used by masses of agents belonging to the computer owners, of whom there are billions and they roam the Internet looking for inefficiencies or things that people are willing to pay for. I'm not sure most people realize how many such projects there are and how large-scale they are.

I didn't really get to test the last Argon twins, but now a new one is out - Carbon. Today it ranks first in most indices among the public models, including a slight advantage over everyone's favorite Opus. Among the interesting things is the Vending Bench benchmark, in which the model won the maximum points. The benchmark basically requires the model to run a business with vendomats for a simulated year (based on the legendary anthropic experiment, which did it physically). The model has to manage supplies, prices, pay rent, transportation to earn as much as possible. This, by the way, hopefully makes it abundantly clear why the "AI will make me money" mission is meaningless nonsense. The better the AI, the worse it will cope with this task, and the more stupid the user, the more often it will ask for it. Reason: the market and capitalism are the first artificial intelligence in the world, which is already several hundred years old. There are no inefficient markets if no one interferes with them. And no one can stop the spread of artificial intelligence and the markets to become more efficient. Therefore, any opportunity for arbitrage or optimization in the market will be used faster than a few years ago. If a trend goes viral on TikTok, then it takes weeks, even months, for that trend to show up in brand advertising, T-shirt prints, and on Mackenzie. It will be minutes. Any "easy" way to make money (even the smallest, let's say a few dollars that a person won't get out of his chair) will immediately be used by masses of agents belonging to the computer owners, of whom there are billions and they roam the Internet looking for inefficiencies or things that people are willing to pay for. I'm not sure most people realize how many such projects there are and how large-scale they are.