Claude 1.0

Claude 1 is a large language model by Anthropic released on March 14, 2023.1

Background

Anthropic claims the “first version” of Claude was trained in spring 2022.2

In November 2022, a 52B parameter model named Anthropic-LM v4-s3 (anthropic/stanford-online-all-v4-s3) was benchmarked.34 This was Anthropic’s largest model on record until a 175B model was benchmarked in February 2023.5

Following the Constitutional AI paper in December 2022,6 Anthropic shared a sign-up form for private testing of models.7 In the first week of 2023, information about Claude began surfacing online from early testers sharing outputs generated through a Slack app.8

Closed beta testing of Claude began in December 2022,7 shortly after the Constitutional AI paper was published.6 In the first week of 2023, public info about Claude began surfacing online. It was initially accessed through a Slack app,8 followed by Quora Poe.9

On March 14, 2023, API early access to Claude was announced.1

Research

Footnotes

  1. Introducing Claude (Anthropic, Mar 14, 2023) ↩ ↩2

  2. Anthropic: Core Views on AI Safety (Anthropic, Mar 9, 2023) ↩

  3. Language Models are Change AI. We Need to Understand Them (Nov 17, 2022) ↩

  4. HELM Classic ↩

  5. The Capacity for Moral Self-Correction in Large Language Models (Ganguli et al., Feb 15, 2023) ↩

  6. Constitutional AI: Harmlessness from AI Feedback (Bai et al., Dec 15, 2022) ↩ ↩2

  7. Anthropic via X (Dec 19, 2022) ↩ ↩2

  8. Anthropic’s Claude improves on ChatGPT but still suffers from limitations (Kyle Wiggers via TechCrunch, Jan 9, 2023) ↩ ↩2

  9. Why choose when you can have both OpenAI and Anthropic? - An evening with Poe - part 1 (Maaike Groenewege, Feb 12, 2023) ↩