The Google-backed startup released an update to its flagship language model

Anthropic, the Google-backed AI startup founded by former OpenAI engineers, has unveiled the next generation of its flagship cloud-based AI product, a large language model. This large language model is called Claude 2, and with its advanced capabilities, it is capable of analyzing hundreds of pages of text and hours of audio. This advanced model has better performance than its predecessor and can provide longer, safer and smarter responses.

According to a blog post from Entropic, Claude 2 has improved performance and can provide longer responses. The company has even released a public beta website to test the model, but for now only US and UK users can access it due to “capacity constraints”. However, this website will be available to all users worldwide in the future.

Claude 2 is based on the latest update of the original Claude model, released in May. This update increased Claude 2’s ability to parse text from 9,000 tokens to 100,000 tokens, which translates into roughly 75,000 words or six hours of transcribed audio per request. This means that this language model is now capable of analyzing larger texts, including business or technical analyzes of long documents such as financial statements or research papers. It can also process longer voice recordings such as podcasts and provide a summary of their content.

For example, Cloud 2 can analyze financial statements by “digitizing” them (converting them into a digital format), summarizing the information in them and explaining their meaning, thus enabling the analysis of strategic risks and opportunities in the financial statements. provided

This model can be used for various purposes such as analyzing the advantages and disadvantages of new laws, reviewing legal and developer documents, and quickly creating or modifying a codebase. Essentially, Anthropic’s Claude can help people save time and make better decisions by automating some of the tasks that would take longer if done manually.

Claude 2 improvements

Additionally, the latest version of Claude has improved coding and math capabilities as well as advanced reasoning.

The developer of Cloud 2 Anthropic has likened this language model to a helpful and supportive colleague or personal assistant who can understand what you want and help you with many tasks. The AI ​​startup offers the Cloud 2 API at no additional cost to businesses using it, i.e. at the same price as its predecessor (Cloud 1.3).

Improved safety in model outputs

Anthropic has researched improving large-scale linguistic models (LLM) with what it calls “legal artificial intelligence.” These include training large linguistic models to answer difficult legal questions while following a set of guiding principles and rules that make these models less vulnerable to attacks. Although no model is completely immune to attacks, Entropic has used various security techniques to improve the accuracy of its model outputs. The company claims to have enhanced the Cloud 2 model with various security techniques, such as red teaming or hiring employees to find weaknesses in the model’s security architecture. In essence, Anthropic is researching ways to train large language models to better answer complex legal questions while improving its security measures to protect against cyber-attacks.

Anthropologists have also experimented with “moral self-correction,” in which language models are trained using feedback from humans to correct themselves and avoid producing harmful outcomes. The researchers believe that this technique could potentially help solve the problem of language models producing outputs that are biased or otherwise harmful.