**Anthropic AI Startup Releases Next Major Model, Claude 2**
Anthropic, an AI startup founded by siblings Dario Amodei and Daniela Amodei, has launched its latest major model, Claude 2. The company claims that Claude 2 has shown significant improvement across key benchmarks such as coding, math, and reasoning skills, while producing fewer harmful answers compared to other AI standouts like OpenAI’s ChatGPT or Inflection’s Pi.
**Improvement Across Multiple Measures**
Claude 2.0 has outperformed its predecessor, Claude 1.3, in various tests. It achieved a score of 71.2% in a Python coding test, up from 56%. Its middle school math quiz grade increased to 88% from 85.2%. Additionally, it raised its Bar exam score to 76.5% from 73%. Furthermore, Claude 2.0 can now handle prompts twice the size of its previous version, allowing for the analysis of more extensive texts like an epic novel.
**Enhanced Availability and Pricing**
Anthropic has made Claude 2.0 more widely available in its second major iteration. The company has launched a beta-test website called claude.ai, allowing general users to register in the U.S. and U.K. Additionally, Anthropic has opened up the new model to businesses via an API at the same price they paid for Claude 1.3. This approach aims to provide a wider safety testing ground and examine potential dangers of the model.
**Commercialization and Business Focus**
Anthropic’s CEO Dario Amodei stated that commercialization was always a part of their plan despite the company’s split from OpenAI over commercialization differences. Amodei believes that commercial adoption allows for a more extensive examination of the model’s safety. The consumer version of Claude 2.0 is currently free, but the company may monetize it in the future.
**Training and Safety Measures**
Anthropic trained Claude 2.0 using a framework called Constitutional AI, which improves the results of AI models without human intervention. However, the model also underwent some level of human feedback and oversight to enhance safety measures. The company claims that Claude 2.0 is twice as effective as its predecessor at limiting harmful outputs. Although perfection is unattainable, Anthropic works to mitigate risks and address potential issues.
**Mitigating AI Risks**
Amodei, along with other industry leaders, has signed a letter expressing concern about the risks of AI leading to human extinction. Anthropic aims to address these risks while continuing to release new models. Instead of supporting a temporary freeze in model releases, the company proposes implementing safety checks and rules for models to ensure responsible development and deployment.
Anthropic’s release of its new flagship model, Claude 2.0, showcases significant improvements in coding, math, and reasoning skills while minimizing harmful outputs. The company has made the model more widely available, both for general users and businesses, and plans to monetize it in the future. Anthropic emphasizes the importance of safety measures and proposes implementing rules and checks to address potential risks associated with AI development and deployment.