OpenAI sparked the global AI revolution with ChatGPT in November 2022. Consequently, we have more people creating new types of AI chatbots, expanding what artificial intelligence can do for us.
Surprisingly, the AI trend enabled competitors to rise against ChatGPT. For the first time, one succeeded, according to the LMSYS Chatbot Arena.
Claude 3 Opus, Anthropic’s latest artificial intelligence, took the top spot on this global leaderboard, pulling down GPT-4 to second place.
So how did Claude beat ChatGPT?
The LMSYS Chatbot Arena pits different AI chatbots against each other by letting them compete in anonymous randomized battles.
READ: How to use Anthropic’s Claude 2 chatbot
It collected more than 400,000 user votes for various models from OpenAI, Google, and Anthropic. Also, it ranks them using the Elo rating system.
Chess.com said it is a widely used ranking system for measuring the relative skill levels of chess players.
The Chatbot Arena included GPT-4, OpenAI’s latest large language model, on May 10, 2023.
It has dominated first place ever since. On March 26, 2024, it dropped to second place due to Anthropic’s Claude 3 Opus model.
Regardless of the LMSYS Chatbot Arena, it is difficult to compare AI chatbot performance objectively.
That is why independent AI researcher Simon Willison told Ars Technica that “vibes” plays an important role in determining LLM quality.
“Vibes” are another term for subjective feelings, and Willison believes Claude’s win is “Yet another case of ‘vibes’ as a key concept in modern AI.”
What is Claude AI?
Most people may not recognize Claude due to ChatGPT’s popularity. According to another Inquirer Tech article, Claude came from the AI firm Anthropic.
OpenAI executives Dario Amodei, Daniela Amodei, Jack Clark, Sam McCandlish, and Tom Brown, founded this company.
Claude recently proved that it can compete against the global leader ChatGPT. However, it has another unique feature called constitutional AI.
It follows 10 principles that serve as its guidelines or “constitution,” which are:
- Beneficence or maximizing positive impact
- Nonmaleficence or avoiding giving harmful advice
- Autonomy or respecting freedom of choice
Anthropic said Claude is less likely to produce harmful or offensive comments than ChatGPT.
You can try Claude by visiting claude.ai for the free version, but it has usage limitations. The site says you may remove those limits by paying $20 monthly plus tax for Claude Pro.