Anthropicthe artificial intelligence company founded by former OpenAI employees, presented Claude 2.1. The new AI model is capable of processing up to 200,000 context tokens, equivalent to 150,000 words or 500 pages of text, which would put it one step ahead of the recently announced GPT-4 Turbo.
The context window is one of the strong points of this new version of the model. Anthropic doubled the amount of information for a simple instruction (prompt), so users can upload documents and complete literary worksas The Iliad and The Odyssey. Claude 2.1 can generate summaries, compare documents, forecast trends, and more in minutes.
“Since our launch earlier this year, millions of people have used Claude for a wide range of applications, from translating academic articles to writing business plans and analyzing complex contracts,” Anthropic mentioned in a blog post. “Processing a 200 KB long message is a complex task and a first for the industry.”
Claude 2.1 has not only doubled the context window, but it is also more precise. Anthropic mentions that this version arrives with significant reductions in hallucination rates.
Here, engineers have worked to minimize the chances of the model giving the wrong answer in open conversations and questions with broad context. After conducting multiple tests, Anthropic found that The percentage of incorrect answers was reduced by 30%while the probability of offering erroneous conclusions is 3 to 4 times lower.
Artificial intelligence too offers improvements in understanding and summarizing long, complex documents. Claude supports legal documentation, financial reports and technical specifications.
With Claude 2.1, Anthropic goes for OpenAI’s jugular
Another novelty of Claude 2.1 is the possibility of integrate with other existing tools and products. The Anthropic model can interact with services, databases, or custom APIs. With this functionality we could connect it to the Internet to carry out a search on the web.
Lastly, Anthropic announced a new feature called system prompts, which provides instructions to Claude before executing a task. The objective of these messages is improve performance through context. We may customize the tone, define style, or ask you to adhere to certain rules prior to the final instruction.
The improvements to the new model will be available to developers through the API, as well as in the chatbot interface. The cost for one million tokens will be 8 dollars, slightly lower price than GPT-4 Turbo. The only detail is that the 200K context window will be exclusive to users subscribed to Claude Pro.
Anthropic is not only in the news for the announcement of its new model, but also because it has emerged that OpenAI sought to merge with its competitor. According to The Informationthe board of directors of OpenAI approached Darío Amodeico-founder and CEO of Anthropic, to unite both companies and become Sam Altman’s replacement.