metamorworks - stock.adobe.com
AI startup Anthropic's newly released generative AI model differentiates itself from competing models such as Bard and ChatGPT by focusing on enterprise and security.
Anthropic -- founded by former members of Microsoft-backed AI research lab and vendor OpenAI -- introduced Claude 2 on July 11.
Claude 2 is the second iteration of Claude, the updated version of its AI assistant based on Anthropic's research.
The AI research lab and vendor had initially allowed only certain partners -- such as Notion, Quora and DuckDuckGo -- to test Claude in a closed alpha. However, in March, it officially introduced the AI assistant to a broader audience by releasing Claude and Claude Instant, a less expensive version. The vendor also incorporated Claude into different applications including Slack and Zoom.
The second generation of the AI assistant has improved performance and more extended responses, according to Anthropic.
Claude 2 can be accessed through an API and a new public-facing website, claude.ai. The AI assistant scored above the 90th percentile on the GRE reading and writing exams, and more than 75% on the multiple-choice section of the bar exam, according to Anthropic. GPT-4 did the same.
The AI assistant is better at giving harmless responses that contain no offensive or inaccurate information compared with the previous generation, according to Anthropic. The API for business is also available at the same price as Claude 1.3, the previous version, and anyone in the U.S. and U.K. can use the beta chat model now.
Users can input up to 100,000 tokens, or about 75,000 words, in each prompt. This means that Claude can quickly access and understand hundreds of pages of technical documents or even a book and write longer papers, according to Anthropic.
How it competes
The ability to handle a larger context window is one of Claude 2's notable differences from other AI assistants and chat tools such as ChatGPT and Bard, Forrester Research analyst Rowan Curran said.
This large context window enables prompts to contain more information, which is crucial for enterprise applications such as natural language query on a structured database, he said.
For this type of use case, the large language model will need to understand the data structure so that it can turn the natural language into a query. Describing the data structure can be challenging, depending on the type of database that is presented, Curran said.
"For businesses adopting genAI and LLMs, it is important to look at these additional factors like the size of the context window, not just how many parameters a model has or how efficiently it answers general-purpose questions," he added.
Another difference is Anthropic's stated commitment to responsible AI, Gartner analyst Arun Chandrasekaran said.
"The strong technical pedigree of Anthropic and its strong focus on responsible and ethical AI can enable it to play an important role [in the generative AI market]," he said.
Mark BeccueAnalyst, Futurum Group
Anthropic's release of Claude 2 also shows the vendor's consideration of the enterprise, said Mark Beccue, analyst at Futurum Group.
"It's further evidence that the generative AI market is mutating," Beccue said. "Here's the interesting thing about that mutation: We are getting to enterprise-grade solutions. Anthropic's design and intent is squarely that."
This chatbot is different from ChatGPT and Bard because those AI assistants were not designed from the beginning as enterprise-grade products.
"Anthropic's market approach is more -- but not entirely -- about leveraging private data sets," Beccue added.
This is important for enterprises drawn to products that meet specific security and reliability standards.
In the past, security and data privacy concerns about ChatGPT have caused some enterprises to ban its internal use. By making Claude 2 an enterprise-grade product, Anthropic wants to remove that barrier.
However, the vendor still warned enterprises that despite using specific training techniques such as constitutional AI -- the practice of using both supervised learning and reinforcement learning to train and control an AI model's behavior -- to improve Claude 2's outputs, no system is tamper-proof.
Meanwhile, Claude 2 provides options within the enterprise and broadens the generative AI market, Beccue said.
Also, since Anthropic is a latecomer to the generative AI market, monetization will be challenging, Chandrasekaran said. To remain competitive, the vendor needs to form stronger partnerships to compete with big tech vendors and other startups such as Cohere, he added.
Anthropic also revealed that Jasper, a generative AI platform, uses Claude 2 for various applications. And the coding platform Sourcegraph uses Claude 2 improved reasoning in its coding assistant Cody.
Esther Ajao is a news writer covering artificial intelligence software and systems.