(Web Desk) - Anthropic, an AI startup supported by Google, asserts that its most recent iteration of GenAI technology, named Claude, outperforms OpenAI's GPT-4 in terms of performance.
According to Anthropic, both Claude and GPT-4 demonstrate heightened capabilities in analysis and forecasting.
The company claims that their AI chatbot excels not only in general performance but also shows enhanced results on specific benchmarks when compared to models like ChatGPT and Google's Gemini 1.0 Ultra (excluding Gemini 1.5 Pro).
Claude 3, Anthropic's latest GenAI iteration, is noteworthy as it represents the company's inaugural foray into multimodal capabilities. This implies that Claude 3 can analyze both text and images, akin to certain versions of GPT-4 and Gemini.
The range of image processing extends to various formats, such as photos, charts, graphs, and technical diagrams sourced from PDFs, slideshows, and other document types.
Distinguishing itself from some GenAI competitors, Claude 3 boasts the ability to analyze multiple images simultaneously, with a maximum limit of 20 in a single request. This feature enables the AI to perform tasks like image comparison and contrast, according to Anthropic.
However, there are constraints on Claude 3's image processing capabilities. Anthropic has deliberately disabled the model's capacity to identify individuals, likely due to ethical and legal considerations.
Additionally, the company acknowledges that Claude 3 may make errors with "low-quality" images (those under 200 pixels) and faces challenges in tasks involving spatial reasoning, such as interpreting an analog clock face, and object counting (the model cannot provide precise counts of objects in images).
“Claude 3 tends to generate more expressive and engaging responses,” Anthropic writes in a support article. “[It’s] easier to prompt and steer compared to our legacy models. Users should find that they can achieve the desired results with shorter and more concise prompts.”
In a detailed technical whitepaper, Anthropic openly acknowledges that Claude 3 faces challenges common to other GenAI models, including issues like bias and hallucinations (generating fictitious information).
Unlike certain GenAI counterparts, Claude 3 lacks the capability to browse the internet; its responses are limited to information available before August 2023. Despite being multilingual, Claude is not as proficient in certain "low-resource" languages when compared to English.
However, Anthropic commits to regular updates for Claude 3 in the coming months, demonstrating their dedication to addressing and improving upon these limitations.
Anthropic aspires to develop a cutting-edge algorithm for "AI self-teaching," envisioning its application in constructing virtual assistants capable of handling tasks like responding to emails, conducting research, and creating various forms of content such as art and books.
This aligns with the capabilities demonstrated by existing models like GPT-4 and other extensive language models that offer a glimpse into the potential of such technology.