AI tech startup Anthropic has launched its latest AI models – a trio named Claude 3, which reportedly outperform solutions from major tech giants. Claude 3 includes three models: Opus, Sonnet, and Haiku, with Opus apparently surpassing AI programs such as OpenAI’s GPT4 and Google’s Gemini 1.0 Ultra in standard capability tests. Sonnet and Haiku are less sophisticated than Opus but while Sonnet and Opus are available in 159 nations, Haiku hasn’t yet been released.

Anthropic co-founder Daniela Amodei asserts that Claude 3 can more effectively handle risk than its predecessor, Claude 2, which was more restrictive in its response. Amodei said that Claude 3 comprehends complex matters better than Claude 2, which was sometimes overly conservative in its replies.

Established in 2021 by ex-OpenAI employees, Anthropic has secured billions in funding, including from Amazon and Google. Unlike its predecessors, Claude 3 enables users to upload documents like snippets, charts, and technical diagrams for the models to analyze. However, generating images is not part of its capabilities.

The Claude 3 models showcase advanced abilities in analyzing, projecting, generating nuanced content, code, and conversing in multiple non-English languages. Despite being more advanced than Claude 2.1, Opus provides results at comparable speeds, with Sonnet operating twice as fast and surpassing Claude 2.1 in intelligence, albeit less capable than Opus.

Additionally, the Claude 3 models are able to deliver citations for users to verify the accuracy of the answers they generate. However, Anthropic acknowledges two primary limitations in Claude 3: hallucinations, errors when the models misjudge visual data, and recognizing harmful images. With the looming 2024 presidential elections, Anthropic is establishing new policies surrounding the use of its AI tools and methodologies to evaluate how the models respond to prompts aimed at election misinformation or other misuses.

Amodei recognizes that no model is perfect and ensures that they’ve diligently tried to make the models as capable and safe as possible.

Source: Move over, ChatGPT: AI startup Anthropic unveils new models that challenge Big Tech }}.