Lazero wants to develop AI systems that are not threat to humans – computers
AI chatbots are not “immune” to black mailAs shown, for example, security and ethics tests Made by anthropic Before releasing the latest version of Claude. In this scenario, Yoshua, which is considered one of the pioneers in the region, led to the launch of a new initiative.
Lazero aims to develop more “honest” AI solutionsThey can detect when the system tries to cheat users. Under the leadership of the researcher, This initiative has approximately 30 million financingAdvance O Guardian.
In solutions that Laziro researchers develop A new system called the scientist AI, which works a kind of security mechanism on AI agents that try to turn against humans.
Unlike the current productive AI tools, the proposed system has no definite answers, but the likelihood of the truth of the response. After being executed with the AI agent, the scientist represents dangerous behaviorsAssessing the probability of their actions causing damage.
Watch the video
The British newspaper quoted Yoshua Bengio, In practice, the scientist acts as a “psychologist” that can understand and assesses harmful behaviors. “We want to develop honest AI systems and do not mislead,” he said.
For the initiative, The first step is to prove that your method is workingOpen and available AI models to receive and train for free. Then Lazero wants to encourage companies and governments to promote a large scale for the development of “honesty” models.
Among the early investors of Lazero Are Future of Life Institute; John TolinOne of the founders of Skype; And the Schmith SciencesA company in the investigation area founded by former Google Executive Director Eric Schmith.
