The startup Anthropic, co-founded by former OpenAI employees and backed by over $700 million in funding, has recently developed a new AI system called Claude that is accessible through a Slack integration. This system appears to improve upon OpenAI’s ChatGPT in several key ways, and has been gaining attention from tech blogs and AI researchers.
Anthropic has built Claude using a technique called “constitutional AI,” which takes a “principle-based” approach to aligning AI systems with human intentions. This approach is based on the concepts of beneficence (maximizing positive impact), nonmaleficence (avoiding giving harmful advice) and autonomy (respecting freedom of choice). Anthropic trained the AI system with a simple set of principles, allowing it to explore possible responses to thousands of prompts and curating those most consistent with the constitution.
Claude is essentially a statistical tool to predict words much like ChatGPT, and can hold an open-ended conversation, tell jokes and provide philosophical insights. AI researchers have pitted Claude against ChatGPT in a battle of wits and have found that Claude generally follows closer to what it’s asked for, answers more trivia questions correctly, is better at telling jokes, and can admit when it doesn’t know the answer to a particularly tough question.
However, Claude isn’t perfect and is susceptible to some of the same flaws as ChatGPT, such as giving answers that aren’t in keeping with its programmed constraints. AI researchers have also found that Claude is worse at math than ChatGPT, does not solve “hallucination” (the AI writing inconsistent, factually wrong statements), and is a poorer programmer.
Overall, Claude appears to be a promising development in AI technology and shows potential for further improvement. Anthropic has not yet made the principles of “constitutional AI” public, but its closed beta of Claude is ongoing and the company plans to refine the system and potentially open the beta to more people down the line. With more testing and refinement, Claude may be a viable alternative to ChatGPT and could open up a world of possibilities for AI applications.