Anthropic debuts ChatGPT’s latest rival
CALIFORNIA, UNITED STATES — Artificial Intelligence (AI) startup Anthropic recently launched Claude 2, an enhanced large language model available, to the public via a beta-test website.
Often compared to OpenAI’s ChatGPT, Claude 2 showed advancements in coding, math, and reasoning skills and produces fewer harmful responses than its predecessor, Claude 1.3.
Anthropic CEO Dario Amodei termed the launch of Claude 2 as an “evolutionary progress,” demonstrating considerable improvement over its forerunner.
In testing, Claude 2 scored higher in Python coding tests, middle school math quizzes, and Bar exams. Additionally, it can handle prompts twice as long as the previous model.
The announcement follows the company’s recent US$450 million funding round led by Spark Capital. Thousands of businesses are already working with Claude’s API, and Anthropic is collaborating with larger customers, such as Zoom and Notion, on customized models.
Despite parting ways with OpenAI over commercialization differences, Amodei acknowledged that commercializing models like Claude 2 was part of Anthropic’s strategy. This commercialization, he stated, provides a broader “safety testing ground” to explore potential risks associated with the model.
Though Claude 2 was trained using “Constitutional AI” and human feedback, Anthropic warns that these models are “never going to be perfect.”
Amodei advocated for rigorous safety checks rather than a temporary freeze on model releases, underscoring the importance of measuring and establishing rules for AI models.
