Anthropic is providing funding for new benchmarks for AI models such as ChatGPT.
Anthropic, an artificial intelligence (AI) company, has launched a new initiative aimed at enhancing how we evaluate AI capabilities. This program, announced on Monday, will provide funding to external organizations to develop improved methods for assessing AI performance.
According to Anthropic, current tests often fall short in reflecting how AI is used in real-world scenarios. Many existing tests were designed before the advent of powerful generative AI and may not accurately measure their capabilities.
To address these shortcomings, Anthropic plans to introduce entirely new testing methods that will be more rigorous and focused. These tests will assess the security implications of AI systems and their potential societal impacts. The company intends to create new tools, resources, and evaluation techniques to achieve this goal.
Anthropic is particularly concerned about the oversight of powerful AI technologies and aims to develop tests that can identify risks associated with them. For instance, these tests will evaluate whether AI models could be exploited for malicious purposes like cyberattacks or the creation of harmful weapons. They will also assess AI’s potential to manipulate public opinion through fake news and other deceptive methods.
In addition to security concerns, Anthropic’s initiative will fund research aimed at harnessing AI for positive applications. This includes exploring AI’s role in scientific research, language translation, and mitigating biases in decision-making processes. Moreover, the company seeks to develop AI models capable of identifying and avoiding the generation of harmful or offensive content.
Anthropic’s overarching goal with this initiative is to advance the field of AI testing to better address emerging challenges and opportunities in AI technology.
