Sat. Apr 13th, 2024

The battle for the best AI language model is raging on. ChatGPT and its connected GPT language model might still be the most popular, but there are a few AI companies trying to do their own thing with varying degrees of success. One of those companies is Anthropic, a company that was started by a few former OpenAI employees and currently makes the “Claude” model. Now, that Claude model has reached version 3, and it’s more scalable and powerful than ever.

Anthropic has just unveiled its new Claude 3 family of AI models claiming they outperform leading models from Google and OpenAI. These models can handle both text and image inputs and offer various improvements over prior versions.

According to Anthropic, Claude 3 boasts better question answering, longer instruction comprehension, and improved accuracy due to its increased context understanding. Three variations exist: Haiku (fastest and most cost-effective), Sonnet (balances speed and performance), and Opus (largest and most intelligent). Each is made to be smaller, or bigger, depending on your own personal needs—Haiku is not as resource-intensive, while Opus is the most capable but also the most resource-intensive out of the three. Regardless, all three variations of Claude 3 are highly versatile and can be used for a wide range of applications, whether that’s to create a chatbot, analyze data, or generate content.

One common complaint about previous iterations of Claude was that it would refuse to comply with requests that actually didn’t violate the model’s code of conduct, but Claude 3 apparently cuts down on that. According to Anthropic, Claude 3 is less prone to such behavior. Additionally, Claude 3 models claim near-instant results even for complex materials like research papers. Benchmarks show Opus surpassing competitors in areas like graduate-level reasoning, math, coding, and general understanding.

Compared to the previous Claude 2.1 model, Sonnet offers double the speed, making it ideal for tasks requiring quick responses. Claude 3 models were trained using a combination of private and public datasets, with training hardware provided by Amazon and Google, both of whom are investors in Anthropic.

These models will, for now, be available on Amazon’s Bedrock model library and Google’s Vertex AI platform.

Source: The Verge

Source link

By John P.

Leave a Reply

Your email address will not be published. Required fields are marked *