Anthropic, a startup that operates in the artificial intelligence industry, has released new software for its chatbot called Claude.

The mentioned company’s new products can handle complex tasks and have a low propensity to fictionalize facts and stories.

On Monday, March 4, a startup based in San Francisco presented three new artificial intelligence models, among which Claude 3 Opus, Sonnet, and Haiku. The AI configuration called Opus has the highest power level. The Haiku artificial intelligence model is the fastest among new digital developments. According to preliminary data, the configuration of Sonnet AI is a kind of intermediate link in the line of virtual products, presented on Monday, in terms of the mentioned parameters.

Opus and Sonnet are already available to developers. Haiku will be offered to consumers in the coming weeks. The relevant information was made public by the startup.

The technology based on which the Claude functioning process is carried out is a large language model that is trained on huge amounts of data stored in the Internet space to generate text materials. The AI can provide information in response to a user’s request or create a literary work on a topic that interests the consumer.

Daniela Amodei, President of Anthropic, says that chatbots tend to generate content containing data that does not correspond to reality. This problem is sometimes characterized as a hallucination. According to Daniela Amodei, the new AI models presented by the startup are still just trained to predict the next word, which is why the risk of hallucinations cannot be eliminated.

At the same time, the company has made efforts to counteract the mentioned problem by creating new configurations of artificial intelligence. The startup said that the digital products presented on Monday are twice as likely to offer consumers the right answers to their questions and are less likely to fictionalize anything. Fixing the hallucination problem is a priority for the clients of Anthropic.

The startup was founded in 2021 by former OpenAI employees, including Daniela Amodei and her brother Dario, who is the chief executive officer. The company is a competitor of OpenAI. The startup raising multibillion-dollar venture investments. The majority of the company’s customer base consists of firms, among which, for example, the DuckDuckGo search engine and the Lonely Planet travel guide publisher.

For Anthropic, one of the main corporate priorities is the safe and responsible development of artificial intelligence. Sometimes the pursuit of this goal has limited the performance of a startup’s digital products. For example, the old versions of the company’s chatbot in some cases refused to provide answers to users’ questions for security reasons, although there was no threat. New AI models are less prone to such incidents.

Dario Amodei says that the science of controlling the training of artificial intelligence systems remains imperfect, but is improving every day.

New AI models from the startup will soon be citing sentences in reference materials to back up the generated responses. This was announced by the company. New artificial intelligence models can analyze images. This means that digital products can perform tasks such as identifying dog breeds in photographs, comparing two images, and describing a piece of art. Similar features are already offered by Alphabet’s Google Gemini and OpenAI’s ChatGPT.

The Anthropic’s models cannot generate images yet. Dario Amodei says that the startup’s clients have not demonstrated the need for an appropriate function.

As we have reported earlier, Microsoft to Launch Copilot AI Chatbot for Finance Workers.