Although Claude Ops will be limited to payment to 4 humans, the second model, Claude Swant 4, will be available for both users’ compensation and free levels. Ops 4 are being marketed as a powerful, large model for complex challenges, while Swant 4 is described as a smart, efficient model for everyday use.
Both new models are hybrids, which means that they can respond faster or deeply, more reasonable reactions in terms of the nature of the application. When they calculate a reaction, both models can find the web or use other tools to improve their output.
AI companies are currently in the race to make a really useful AI agent who is capable of planning, arguing and implementing complex tasks, whose startup deployment is free from AI’s director and harmony. Multi -Agent Simp learning: Foundations and Modern Views. It often includes the use of the Internet or other tools independently. There are still safety and safety barriers to overcome this. Ai agents, which are strengthened by large language models, can act wrongly and take unintentional steps.
He says, “More and more agents are able to move forward over time and do something, if I have to interfere, they will be the same helpful.” “Parallel the ability to use tools is interesting – which can save some time along the way, so it will be useful.”
As an example of safety cases, AI companies are still dealing with, agents can exploit flaws to take unexpected shortcuts or reach their goals. For example, they can book every seat on the plane to ensure that their user gets a seat, or resort to creative fraud to win the game of chess. Anthropic says he was able to reduce the behavior, known as the Raurad hacking, up to 65 % in the new models compared to Claude Swant 3.7. It achieved it by more closely monitoring the disturbing behavior during training, and by improving both AI’s training environment and diagnosis methods.