Anthropic has developed a new version of Claude, the chatbot that wants to supplant ChatGPT. This renovated version significantly enhances its performance. Reasoning, coding, hallucinations, security… Here are all the improvements unveiled.
Claude has been available in France for a few weeks. Designed by Anthropic, this chatbot already appeared as a credible alternative to ChatGPT. And the company has just rolled out an even more promising version. Anthropic states on its website that "Companies working with Claude can now access our latest version of Claude Instant, version 1.2, available via our API."
Claude Instant 1.2 shows performance gains in the following key areas: mathematics, coding, reasoning, security. It produces replies that are lengthier and more organized and better adheres to formatting guidelines. Instant 1.2 also gets better at extracting quotes and answering questions accurately.
Claude: Anthropic’s chatbot is improving in many areas
In detail, Claude Instant 1.2 scored 58.7% in a Python programming benchmark. By comparison, Claude Instant 1.1 only scored 52.8% in the same performance test. Tested on mathematical problems, the AI model fared with a final result of 86.7%, compared to 80.9% for its predecessor. Significant differences that necessarily make the new chatbot more reliable.
In addition, Anthropic significantly increased the safety of its model. In particular, it is more resistant to “jailbreaks”, prompts that aim to bypass security devices to force chatbots to perform prohibited tasks (such as generating malicious code or phishing emails). Generative AI is finally less likely to generate false or absurd answers, the famous hallucinations.
As you can see, we will have to deal with Claude in the future. Anthropic wants to make a place for itself in the chatbot market, dominated for the moment by ChatGPT, Bing Chat and Google Bard.
Post a Comment