Screenshot by Lance Whitney/ZDNET.
Attention, ChatGPT, a new AI chatbot is intruding on your territory!
Released Monday, the third version of Anthropic’s Claude AI is said to be more competent, better informed, and better at reasoning than OpenAI’s ChatGPT and Google’s Gemini.
Opus, Sonnet and Haiku
Claude 3 is a unique product that offers three different models. Opus and Sonnet are already available via the Claude 3 website and as an API for developers. The faster Haiku model will be available soon, according to Anthropic.
According to Anthropic’s research, the Opus model outperforms GPT-3.5, GPT-4, and Gemini in several key areas. The company’s tests covered general knowledge, but also university-level knowledge – including expert reasoning – basic math problem solving and coding skills. Thanks to his training and more in-depth knowledge, Claude 3 presents “levels of understanding and fluency close to those of humans for complex tasks,” argues Anthropic.
Claude 3 also boasts much faster response times. The Sonnet model, in particular, is twice as fast as the Claude 2 and Claude 2.1 versions, according to tests. This model is ideal, according to Anthropic, for searching for information or automating sales.
Haiku is the fastest of the three models. He is able to read a dense research paper with tables and illustrations in less than three seconds.
More understanding and less hallucinations
Anthropic also claims that Claude 3 is more accurate and less error-prone than its previous versions. To verify this claim, the company subjected its various models to a large number of complex and factual questions. With the Opus model, Claude 3 obtained twice as many correct answers as Claude 2.1. The new version also produced fewer wrong answers and hallucinations.
To avoid providing harmful information, AIs often refuse to answer questions deemed inappropriate. But it also happens that they mistakenly interpret a harmless question as harmful. In Anthropic’s testing, Claude 3 was less likely than its previous versions to refuse to answer innocuous questions. In this regard, the three Claude 3 models demonstrated a better understanding of queries and a greater ability to distinguish harmful from harmless questions.
Anthropic also touts Claude 3 as being easier to use, able to accept longer messages and better retain information from previous messages.
Analyze and summarize documents
Among Claude’s improvements, one of the main ones is its support for files in queries. It is now possible to submit various types of files to the AI, including images, PDFs, texts, Microsoft Office files, spreadsheets in CSV format or even HTML files.
Depending on what you ask him, Claude will then be able to analyze, summarize and answer questions about the content of these files.
To test Claude 3, visit Anthropic’s AI website. The free version of the site uses the Sonnet template to answer your questions. The Claude Pro subscription, available for $20 per month, uses the Opus model, the most advanced. It also allows several advantages, including priority access when there is a peak in use and early access to new features.
Source: ZDNet.com