Anthropic, the AI research lab, has recently launched Claude 2, a chatbot that rivals OpenAI’s ChatGPT. The new language model is designed to excel in a wide range of tasks, including coding, text analysis, and composition writing. Unlike the previous version, Claude 2 is now available for open beta testing on a new website, free of charge for users in the US and UK. Additionally, developers can access the model through the commercial API.
Claude 2 aims to simulate a conversation with a helpful colleague or personal assistant. Anthropic has incorporated feedback from users to enhance the model and make it more reliable. According to the company, users have found Claude 2 to be easy to converse with, capable of explaining its thought process and less prone to producing harmful outputs. Additionally, the improved model possesses a longer memory.
The advancements made in Claude 2 are visible in three key areas: coding, math, and reasoning. According to Anthropic, the latest model demonstrated notable improvements in coding proficiency, scoring 71.2% on the Codex HumanEval Python programming test, up from 56.0% with the previous version. Furthermore, its performance on the GSM8k math problem test increased from 85.2% to 88.0%.
A major breakthrough with Claude 2 is its ability to handle longer input and output lengths. Anthropic has pushed the limits, experimenting with prompts consisting of up to 100,000 tokens. This allows the model to analyze lengthy documents, such as technical guides and even entire books. Notably, the expanded output length allows for the generation of longer documents as well.
The new model addresses concerns about generating harmful or offensive outputs. Anthropic conducted an internal evaluation and found that Claude 2 was twice as effective in providing harmless responses compared to its predecessor, Claude 1.3.
Claude 2 is not limited to individual users but is also available to businesses through its API. Companies such as Jasper, an AI writing platform, and Sourcegraph, a code navigation tool, have already started integrating Claude 2 into their operations.
Despite its advancements, Anthropic acknowledges that language models like Claude 2 have limitations. While they can analyze complex works, it is important to exercise caution and not rely on them as factual references. Anthropic advises users to leverage the model for tasks like summarizing or organizing information, while being familiar with the subject matter and validating the results.
In conclusion, Anthropic’s new offering, Claude 2, presents an AI chatbot that competes with the likes of ChatGPT. With its enhanced abilities in coding, math, and reasoning, alongside improved user feedback and reduced generation of harmful outputs, Claude 2 is poised to make an impact in both individual and business settings. As AI technology continues to evolve, it is crucial to understand the limitations of these models and utilize them accordingly.
