Jump to content

Claude (language model)

From Wikipedia, the free encyclopedia
(Redirected from Constitutional AI)

Claude
Developer(s)Anthropic
Initial releaseMarch 2023; 1 year ago (2023-03)
Type
LicenseProprietary
Websiteclaude.ai

Claude is a family of large language models developed by Anthropic.[1][2] The first model was released in March 2023.

The Claude 3 family, released in March 2024, consists of three models: Haiku optimized for speed, Sonnet balancing capabilities and performance, and Opus designed for complex reasoning tasks. These models can process both text and images, with Claude 3 Opus demonstrating enhanced capabilities in areas like mathematics, programming, and logical reasoning compared to previous versions.[3]

Training

[edit]

Claude models are generative pre-trained transformers. They have been pre-trained to predict the next word in large amounts of text. Then, they have been fine-tuned, notably using constitutional AI and reinforcement learning from human feedback (RLHF).[4][5]

Constitutional AI

[edit]

Constitutional AI is an approach developed by Anthropic for training AI systems, particularly language models like Claude, to be harmless and helpful without relying on extensive human feedback.[6] The method, detailed in the paper "Constitutional AI: Harmlessness from AI Feedback" involves two phases: supervised learning and reinforcement learning.[7][8]

In the supervised learning phase, the model generates responses to prompts, self-critiques these responses based on a set of guiding principles (a "constitution"), and revises the responses. Then the model is fine-tuned on these revised responses.[8]

For the reinforcement learning from AI feedback (RLAIF) phase, responses are generated, and an AI compares their compliance with the constitution. This dataset of AI feedback is used to train a preference model that evaluates responses based on how much they satisfy the constitution. Claude is then fine-tuned to align with this preference model. This technique is similar to RLHF, except that the comparisons used to train the preference model are AI-generated, and that they are based on the constitution.[9][6]

The "constitution" for Claude included 75 points, including sections from the UN Universal Declaration of Human Rights.[7][4]

Models

[edit]
Screenshot of an example of Claude 3.5 Haiku answer describing Wikipedia

The name Claude was notably inspired by Claude Shannon, a pioneer in artificial intelligence.[10]

Claude

[edit]

Claude was the initial version of Anthropic's language model released in March 2023,[11] Claude demonstrated proficiency in various tasks but had certain limitations in coding, math, and reasoning capabilities.[12] Anthropic partnered with companies like Notion (productivity software) and Quora (to help develop the Poe chatbot).[12]

Claude Instant

[edit]

Claude was released as two versions, Claude and Claude Instant, with Claude Instant being a faster, less expensive, and lighter version. Claude Instant has an input context length of 100,000 tokens (which corresponds to around 75,000 words).[13]

Claude 2

[edit]

Claude 2 was the next major iteration of Claude, which was released in July 2023 and available to the general public, whereas the Claude 1 was only available to selected users approved by Anthropic.[14]

Claude 2 expanded its context window from 9,000 tokens to 100,000 tokens.[11] Features included the ability to upload PDFs and other documents that enables Claude to read, summarize, and assist with tasks.

Claude 2.1

[edit]

Claude 2.1 doubled the number of tokens that the chatbot could handle, increasing it to a window of 200,000 tokens, which equals around 500 pages of written material.[15]

Anthropic states that the new model is less likely to produce false statements compared to its predecessors.[16]

Claude 3

[edit]

Claude 3 was released on March 14, 2024, with claims in the press release to have set new industry benchmarks across a wide range of cognitive tasks. The Claude 3 family includes three state-of-the-art models in ascending order of capability: Haiku, Sonnet, and Opus. The default version of Claude 3, Opus, has a context window of 200,000 tokens, but this is being expanded to 1 million for specific use cases.[17][3]

Claude 3 drew attention for demonstrating an apparent ability to realize it is being artificially tested during needle in a haystack tests.[18]

Claude 3.5

[edit]
Example of Claude 3.5 Sonnet output

On June 20, 2024, Anthropic released Claude 3.5 Sonnet, which demonstrated significantly improved performance on benchmarks compared to the larger Claude 3 Opus, notably in areas such as coding, multistep workflows, chart interpretation, and text extraction from images. Released alongside 3.5 Sonnet was the new Artifacts capability in which Claude was able to create code in a dedicated window in the interface and preview the rendered output in real time, such as SVG graphics or websites.[19]

An "upgraded Claude 3.5 Sonnet" was introduced on October 22, 2024, along with Claude 3.5 Haiku. A feature, "computer use," was also unveiled in public beta. This capability enables Claude 3.5 Sonnet to interact with a computer's desktop environment, performing tasks such as moving the cursor, clicking buttons, and typing text, effectively mimicking human computer interactions. This development allows the AI to autonomously execute complex, multi-step tasks across various applications.[20]

Criticism

[edit]

Claude 2 received criticism for its stringent ethical alignment that may reduce usability and performance. Users have been refused assistance with benign requests, for example with the programming question "How can I kill all python processes in my ubuntu server?" This has led to a debate over the "alignment tax" (the cost of ensuring an AI system is aligned) in AI development, with discussions centered on balancing ethical considerations and practical functionality. Critics argued for user autonomy and effectiveness, while proponents stressed the importance of ethical AI.[21][16]

References

[edit]
  1. ^ "What is Claude AI?". IBM.
  2. ^ Henshall, Will (July 18, 2023). "What to Know About Claude 2, Anthropic's Rival to ChatGPT". TIME. Retrieved December 14, 2024.
  3. ^ a b Whitney, Lance (March 4, 2024). "Anthropic's Claude 3 chatbot claims to outperform ChatGPT, Gemini". ZDNET. Retrieved March 5, 2024.
  4. ^ a b "What to Know About Claude 2, Anthropic's Rival to ChatGPT". TIME. July 18, 2023. Retrieved January 23, 2024.
  5. ^ Nuñez, Michael (May 9, 2023). "Anthropic releases AI constitution to promote ethical behavior and development". VentureBeat. Retrieved November 17, 2024.
  6. ^ a b Edwards, Benj (May 9, 2023). "AI gains "values" with Anthropic's new Constitutional AI chatbot approach". Ars Technica. Retrieved November 17, 2024.
  7. ^ a b Bai, Yuntao; Kadavath, Saurav; Kundu, Sandipan; Askell, Amanda; Kernion, Jackson; Jones, Andy; Chen, Anna; Goldie, Anna; Mirhoseini, Azalia (December 15, 2022), Constitutional AI: Harmlessness from AI Feedback, arXiv:2212.08073
  8. ^ a b "Claude's Constitution". Anthropic. May 9, 2023. Retrieved March 26, 2024.
  9. ^ Eliot, Lance (May 25, 2023). "Latest Generative AI Boldly Labeled As Constitutional AI Such As Claude By Anthropic Has Heart In The Right Place, Says AI Ethics And AI Law". Forbes. Retrieved March 27, 2024.
  10. ^ Roose, Kevin (July 11, 2023). "Inside the White-Hot Center of A.I. Doomerism". The New York Times.
  11. ^ a b Drapkin, Aaron (October 27, 2023). "What Is Claude AI and Anthropic? ChatGPT's Rival Explained". Tech.co. Retrieved January 23, 2024.
  12. ^ a b "Introducing Claude". Anthropic. March 14, 2023.
  13. ^ Yao, Deborah (August 11, 2023). "Anthropic's Claude Instant: A Smaller, Faster and Cheaper Language Model". AI Business.
  14. ^ Matthews, Dylan (July 17, 2023). "The $1 billion gamble to ensure AI doesn't destroy humanity". Vox. Retrieved January 23, 2024.
  15. ^ Davis, Wes (November 21, 2023). "OpenAI rival Anthropic makes its Claude chatbot even more useful". The Verge. Retrieved January 23, 2024.
  16. ^ a b "Anthropic Announces Claude 2.1 LLM with Wider Context Window and Support for AI Tools". InfoQ. Retrieved January 23, 2024.
  17. ^ "Introducing the next generation of Claude". Anthropic. Retrieved March 4, 2024.
  18. ^ Edwards, Benj (March 5, 2024). "Anthropic's Claude 3 causes stir by seeming to realize when it was being tested". Ars Technica. Retrieved March 9, 2024.
  19. ^ Pierce, David (June 20, 2024). "Anthropic has a fast new AI model — and a clever new way to interact with chatbots". The Verge. Retrieved June 20, 2024.
  20. ^ "Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku". www.anthropic.com. Retrieved October 25, 2024.
  21. ^ Glifton, Gerald (January 3, 2024). "Criticisms Arise Over Claude AI's Strict Ethical Protocols Limiting User Assistance". Light Square. Retrieved January 23, 2024.
[edit]
Retrieved from "https://en.wikipedia.org/w/index.php?title=Claude_(language_model)&oldid=1264824611#Constitutional_AI"