Claude

Anthropic, an AI startup co-founded by former employees of OpenAI, has quietly begun testing a new AI assistant named Claude. As part of our early access to the tool we have been able to compare it against ChatGPT and share some initial findings. Claude does not provide many technical details; however Anthropic’s research paper on Constitutional AI describes their 52-billion-parameter pre-trained model called AnthropicLM v4-s3 which was trained unsupervised on a large text corpus similar to OpenAI’s GPT-3. Anthropic states that Claude is a larger model with architectural choices based on the published research. Both Claude and ChatGPT use reinforcement learning (RL) for training preference models over their outputs, but the methods used for developing these preference models differ with Anthropic favoring an approach they call Constitutional AI.