Anthropic recently announced an upgrade to their conversational AI assistant Claude, releasing Claude 2.1. This latest version makes the already powerful AI chatbot Claude more useful and reliable for enterprise and individual use cases.
In this article, we’ll go over some of the key changes and what’s been added along with this new update.
200K Context Window
One of the biggest changes in Claude 2.1 is the increased context window size. Previously capped at 100,000 tokens, Claude can now process queries and documents with up to 200,000 tokens of context! That translates to about 150,000 words or 500+ pages of text.
This expanded context window enables Claude to handle long, complex documents like legal briefs, financial reports, academic papers, and even literary works. Users can upload entire codebases, SEC filings, or books for Claude to summarize, analyze, and answer questions about.
Claude has always been good at processing information – especially how it automatically creates a .txt file when you copy and paste data in. Now it appears as though it’s going to be even better at processing information, thanks to the increased limit!
Reduced Hallucination Rates
One of the main complaints users had about Claude (and most AI assistants in general) are the out of place hallucinations they sometimes have. This is when the AI tends to go off on a tangent or give a completely unrelated or inaccurate response.
In addition to the larger context capacity, Claude 2.1 boasts a 2x reduction in hallucinated content or false claims compared to the previous version. Through curated testing, Anthropic found Claude 2.1 will more likely admit uncertainty than provide an incorrect answer.
This improved honesty and accuracy makes Claude better suited for business use cases that demand reliable and precise outputs.
New API Beta Feature
One exciting new capability introduced in Claude 2.1 is the API Tool Use. This feature allows Claude to integrate with a user’s existing systems and external tools to expand what it can accomplish.
In simple terms, tool use enables Claude to “plug into” developer-created functions, APIs, databases, and more to take actions on a user’s behalf. Developers can define and register these external tools, data sources, and functionality. Then they provide Claude with instructions that leverage those connected tools to achieve tasks.
For example, here’s some of the things that Anthropic has suggested developers could link Claude to:
- A calculator API for complex math and numerical analysis
- Translation services to interpret text or speech in other languages
- Private company knowledge bases to enable Claude to search and retrieve answers from document stores or databases
- Internal software and workflows via custom APIs, allowing Claude to take basic programmed actions to assist end users
On top of those major changes, Claude 2.1 boasts better comprehension and summarization abilities, especially for lengthy technical material. Anthropic also updated the developer console to make it easier to test and optimize prompts.
Specifically, Claude 2.1 showed a 30% decrease in incorrect answers when queried about details and claims in these long-form materials. Anthropic has also introduced system prompts, which allow developers to provide Claude with custom instructions for how to structure and frame responses. This helps clamp down on unhelpful or unnecessary output and can tailor Claude’s “personality” for specific use case needs.
Claude 2.1 is available now through Anthropic’s API and powers the main claude.ai chat tool. Unfortunately, for now, the 200k context limit is limited to paying Claude Pro users. That said, Claude’s base limits are already extremely generous, and 200k is likely not going to be needed by the average person.
With significant improvements in accuracy, honesty, and versatility, Claude 2.1 represents an exciting upgrade for Anthropic’s AI assistant. We can expect more enterprises to leverage Claude for complex business needs as trust and capabilities continue advancing.
OpenAI’s ChatGPT may be innovating in other fields such as with their custom GPTs, but with these latest improvements, Claude 2.1 still definitely has its uses. We’ve found that as a simple browser assistant (i.e. letting the AI read and analyze your web pages), Claude works fantastic in most cases.