Anthropic Releases Faster, Cheaper Text-Generating Model
Anthropic, the AI startup co-founded by ex-OpenAI execs, has released an updated version of its text-generating model called Claude Instant 1.2.
The new version incorporates improvements from Anthropic’s flagship model, Claude 2, showing significant gains in math, coding, reasoning, and safety.
In internal testing, Claude Instant 1.2 scored higher compared to previous versions in coding benchmarks and math questions.
According to Anthropic, Claude Instant 1.2 also generates longer, more structured responses, follows formatting instructions better, and demonstrates better quote extraction, multilingual capabilities, and question answering.
Anthropic claims that Claude Instant 1.2 is less likely to generate incorrect or nonsensical text (hallucination) and is more resistant to jailbreaking attempts.
Claude Instant 1.2 has a context window size of 100,000 tokens, similar to Claude 2, allowing them to analyze about 75,000 words.
Large context windows help models remember recent conversations without forgetting important content.
While Anthropic aims to create a next-gen algorithm for AI self-teaching, Claude Instant is designed to compete with similar entry-level offerings from OpenAI, Cohere, AI21 Labs, and others.
Anthropic has raised $1.45 billion to date but estimates it will need $5 billion over the next two years to develop its envisioned chatbot.
Anthropic’s Customers and Partners
Anthropic currently has “thousands” of customers and partners, including Quora, which offers access to Claude and Claude Instant through its generative AI app Poe.
Claude powers DuckDuckGo’s DuckAssist tool, which provides direct answers to simple search queries in combination with OpenAI’s ChatGPT.
On Notion, Claude is integrated into the technical backend for Notion AI, an AI writing assistant within the Notion workspace.