The competitive landscape of artificial intelligence is welcoming a new participant. While tech giants like Microsoft-supported OpenAI and Google are vying for supremacy in developing the most progressive AI chatbots, Anthropic is also making significant investments to stay competitive.
Only a few monnths after securing $750 million in two funding rounds, the startup is unveiling a fresh AI chatbot: Claude 2.
Established in 2021 by previous OpenAI research executives, Anthropic has drawn financial backing from companies like Google, Salesforce, and Zoom. With the release of Claude 2, the startup is offering its chatbot technology to the general public for the first time. Over the last two months, Anthropic's AI models have undergone testing by businesses such as Slack, Notion, and Quora, amassing a waitlist of over 350,000 people eager to access Claude's API and consumer services.
"Our efforts have primarily been directed towards enterprises, aiming to maximize Claude's safety," expressed Daniela Amodei, Anthropic's co-founder, alongside her brother, Dario. "We believe that this iteration of Claude is the safest we've created so far, thus we're eager to share it with a wider demographic of both businesses and individual consumers."
To begin, Claude 2 will be accessible only to users in the U.S. and U.K., with plans for broader availability in the upcoming months.
The introduction of ChatGPT by OpenAI last year spurred heavy investment in generative AI chatbots, capable of responding to text prompts with detailed and conversational responses. Despite significant concerns about the potential for perpetuating bias, the technology has rapidly penetrated various sectors including education, online travel, healthcare, online advertising, and more.
In March, OpenAI launched GPT-4, a major upgrade to ChatGPT's foundational technology. In May, the chatbot was given internet browsing capabilities to overcome its 2021 training data cut-off limitation, but the feature was later withdrawn due to reports of misuse, like evading paywalls.
In February, Google revealed its Bard competitor, which has since been enhanced with improved math and coding capabilities and incorporated into more products.
Anthropic has similarly grand objectives.
Claude 2 boasts the ability to summarize roughly 75,000 words, equivalent to the length of a book. Users can input extensive data sets and request summaries as a memo, letter, or story. In comparison, ChatGPT can manage about 3,000 words.
Anthropic spent at least two months crafting Claude 2, with a dedicated team of 30-35 people working on the AI model and an overall support team of 150 individuals. Amodei noted that the rapid expansion of the market provides ample opportunities for multiple players to thrive.
During May, Anthropic joined Alphabet, Microsoft, and OpenAI as one of the four companies invited to a White House meeting on responsible AI development, hosted by Vice President Kamala Harris. The company also raised $450 million in the same month, following a $300 million funding round in March, pushing its valuation to $4.1 billion.
Amodei illustrated the advancements of Claude 2 over its predecessor with tangible examples. The updated chatbot achieved 76.5% on the Bar exam's multiple choice section, a rise from 73%. In a Python coding test, Claude 2's score increased to 71% from the previous 56%.
Concerning safety, Claude 2 was twice as effective in generating "harmless responses", according to a company blog post. However, it's worth noting that the safety evaluation was conducted in-house.
Regardless of the improvements of Claude 2, Amodei acknowledged the many forthcoming challenges for Anthropic and the entire industry. The persistent issue of AI chatbots creating inaccurate responses, termed as "hallucinations" by some tech firms, remains unsolved.
Amodei stated, "There are numerous known and unknown challenges with all language models today. No language model is entirely immune from hallucinations, and Claude 2 is no exception."