11-7-2023 (CALIFORNIA) An artificial intelligence startup backed by Google has widened public access to its chat program Claude and upgraded its AI technology that it says makes “Claude 2” better suited for tasks like computer coding and math.
The improvements come just months after Claude first launched, showing how quickly companies seek advantages in AI. Other players include Microsoft, Inflection AI and Google itself.
Claude 2 promises to substantially speed up reading and writing tasks. It can draft blog posts or analyze existing text.
Anthropic aims to stand out by Claude reading around 75,000 words at once, letting it pore over long documents.
Interest in regulating AI has grown due to copyright concerns and AI generating harmful content. Anthropic says Claude 2 doubles its safety performance.
Rival OpenAI’s ChatGPT AI responds to images in addition to text, unlike Claude. It scored 75.7% on a multiple choice bar exam, a test for lawyers.
Anthropic says Claude 2 now scores 76.5% on that section, up from 73%. A spokesperson did not compare that directly to OpenAI’s result.
Internally, Anthropic discusses multimodal systems and leveraging Claude’s ability to quickly understand vast amounts of content, says Sandy Banerjee.
Anthropic’s models “tend to be some of the faster ones, so for quick customer decisions, Claude is compelling”, Banerjee said.