OpenAI in a crisis while Anthropic releases an improved Claude chatbot

• Anthropic introduces a new chatbot update, Claude 2.1.
• Anthropic Claude 2.1 introduces advanced features like enhanced text processing and truthful response generation. 
• The latest update of Anthropic Claude 2.1 showcases innovative ‘tool use’ capabilities, setting new industry standards.

OpenAI is currently busy managing – and then apparently re-managing – its organizational changes, including leadership roles. Meanwhile Anthropic, an AI startup supported by Google, has released a new chatbot update. This update introduces Claude 2.1, designed for pro users and capable of handling up to 200,000 tokens, equivalent to over 500 content pages.

Anthropic claims that Claude 2.1 can process more text and generate more accurate responses than its predecessors. Additionally, it can integrate with developer-specific APIs, enhancing its compatibility with various technological infrastructures.

Originally established in 2019 by former OpenAI employees focusing on machine learning safety and reliability, Anthropic has enhanced Claude 2.1’s capabilities. This version powers the company’s web-based AI chatbot application and is accessible to developers and enterprises. Similar to OpenAI’s ChatGPT, Claude accepts natural language prompts and attempts to provide answers through conversations.

In its latest version, Claude 2.1 offers improved business features, including an unprecedented 200K token context window, reduced hallucination rates, and a new beta feature for tool use, as noted in Anthropic’s release.

The token context window, now capable of handling double the tokens compared to Claude 2, sets a new industry standard. This enhancement means Claude can hold about 150,000 words or over 500 pages of text, allowing for more extensive natural language tasks like summarization and translating complex documents. However, processing such large amounts of text may take the chatbot several minutes.

Anthropic Claude 2.1’s commitment to truthful responses

One significant improvement in Claude 2.1 is its enhanced truthfulness in responses. According to Anthropic, this model fabricates information at a rate twice as low as its predecessor. Additionally, unlike some competing systems, it tends to acknowledge when it doesn’t know an answer rather than inventing one.

To assess Claude 2.1’s honesty, Anthropic conducted tests using complex, factual questions to challenge the model’s accuracy. The company employed a method to differentiate between false statements and acknowledgments of uncertainty. Results showed that Claude 2.1 was more likely to express uncertainty than to provide wrong information.

Similar capabilities are found in other chatbots, such as Google Bard, which verifies its responses against search results and differentiates between verified facts and dubious claims.

An impressive achievement of Claude 2.1 is its 30% decrease in incorrect responses and a significantly lower tendency to wrongly interpret a document as supporting a specific assertion.

Innovative ‘tool use’ feature

Responding to user demand, the company introduced a new beta feature called ‘tool use,’ letting Claude integrate with existing user processes, products, and APIs. This feature is designed to enhance Claude’s practicality in daily operations.

Claude can now handle tasks across various developer-defined functions or APIs, search the web, and access information from private databases. Users can set up a range of tools for Claude to use and specify tasks, with the model determining the necessary device and executing the job, such as performing complex calculations, converting natural language into API calls, searching databases or the web, executing simple software actions, and connecting to product databases for recommendations and purchase assistance.

This tool use feature is in its initial stages, with development focused on creating features and guidelines to facilitate application integration.

For developers, Anthropic is streamlining the developer console for Claude API users, making it easier to test and learn from new prompts. The new workbench product offers a playground-like environment for prompt iteration and access to settings to fine-tune Claude’s functionality.

Developers can create, navigate, and save multiple prompts for various projects, retaining historical context. They can also generate code snippets to integrate prompts directly into SDKs.

Additionally, the company is introducing system prompts, which let users give Claude custom instructions to enhance performance. These prompts help Claude adopt specific personalities and roles, and structure responses more effectively according to user needs.

Pricing and accessibility for users

Claude 2.1 is now available through its API and powers the chat interface at claude.ai for free and pro users. The 200K token context window feature is exclusive to pro users, allowing for the upload of larger files. The company is excited to see how these new features will inspire use cases as it aims to develop the safest and most advanced AI systems.

The pricing for Claude 2.1 usage is set at US$8 per million tokens for input prompts and US$24 per million tokens for model-generated output.

While this release was happening, The Information reported that OpenAI’s board was considering replacing CEO Sam Altman with Anthropic’s CEO Dario Amodei, who declined both the offer and a proposed merger. Following Altman’s departure, the board offered the interim CEO position to ex-Twitch CEO Emmett Shear, who accepted. Meanwhile, Altman joined Microsoft along with other OpenAI personnel. This leadership change led to a near-mass resignation threat from OpenAI’s employees, and at last call, Altman had returned to the CEO role at OpenAI.

Anthropic, founded by former OpenAI executives who left due to disagreements on AI safety and governance, has received investments from Google’s Alphabet and Amazon.com. Its Claude AI models compete with OpenAI’s GPT series for prominence in the AI industry.

Arguably, Anthropic has had a week showing the difference between an AI company that’s continually improving its model, and one that’s being consumed in personality conflicts and just occasionally absurdity. And both Anthropic and Claude have come out on top.

---

---

---