by Anthropic, the company founded by former OpenAI employees, has launched this Wednesday the new chatbot Claude, which offers functions very similar to those of ChatGPT, but is “more manageable” and shows “less harmful results”.
The Artificial Intelligence (AI) company has advanced that Claude is capable of receiving instructions to perform tasks such as searching for documents, summarizing information, writing text naturally and answering questions on specific issues. The user can even give instructions on personality, tone and desired behavior.
However, the difference with respect to other models such as ChatGPT is that it is a chatbot with qualities that make it “easier to talk to and more manageable”being, in addition, “much less likely to produce harmful results”, as the company has stated in a statement on its blog.
In this sense, Anthropic launched Claude on Tuesday, making two versions available to users through an application programming interface (API). These are Claude-v1, the high-performance and latest-generation option, and Claude Instant, the most economical model.
Specifically, as reflected by the company, prices revolve around $1.45 (1.37 $) per million characters in the case of Claude Instant and $8.60 (8.15 $) per million characters. if using Claude-v1.
Like ChatGPT, Claude does not have internet access and has been trained on data from public web pages until spring 2021. Thus, according to the statements of an Anthropic spokesperson collected by TechCrunch, this chatbot was “trained to avoid sexist, racist and toxic responses”, as well as “to avoid helping a human being to participate in illegal or unethical activities”.
In this sense, What sets Claude apart is a technique called “constitutional AI,” says the developer company. To do this, Anthropic includes a list of around ten principles that, together, make up a kind of constitution (hence the name “constitutional AI”).
The company has not made public the specific principles on which this technique is based. However, they assure that they are inspired by charity, that is, in maximizing the positive impact; non-maleficence, which refers to avoiding giving harmful advice to the user; and autonomy, which consists of respecting freedom of choice.
Anthropic has released Claude after testing with partners like Notion, Quora, and DuckDuckGo. Now, your API has been released globally and you have already made an early access link available. However, the spokesperson has stressed that they are not looking for user-level clients at the moment, but rather “start-ups making bold technological bets”, in addition to “larger and more established companies”. In this way, the company intends to be able to offer a “superior and specific” product.
The evolution of ChatGPT appears
OpenAI has officially introduced the new generation of its language model, GPT 4, designed to accurately solve large problems and provide more useful and secure answers, with the ability to generate from text and image inputs.
GPT 4 stands out for its creativity and ability to collaborate. In addition to generating original content, you can also edit and repeat tasks with users, such as composing songs or scripts, also learning their style.
If one of the premises of the previous models was the generation of content from text, now you can compose a subtitle, carry out an analysis or a classification from an image. youIt is also capable of managing longer text, of more than 25,000 words.
*With information from Europa Press.
