Anthropic Unveils New AI Models Claiming Improved Reasoning and Safety
-
Anthropic unveils new Claude 3 models that it says outperform models from OpenAI and Google in benchmarks of AI capabilities like reasoning.
-
The Claude 3 suite contains 3 models - Opus, Sonnet, and Haiku - with Opus being the most capable. Sonnet prioritizes speed over intelligence.
-
Claude 3 adds capabilities like analyzing graphs/images and translating languages compared to the previous Claude 2 model.
-
However, Claude 3 still sometimes hallucinates or fails to recognize harmful images, requiring more safety development.
-
With 2024 elections nearing, Anthropic aims to develop policies against potential misuse of models for generating political misinformation.