Decoder. plain-English AI glossary

Claude

● Core

Anthropic's line of language models, including Claude 1, 3.5 Sonnet, and Opus; built with alignment and safety as first-class concerns.

Think of it like

The safety-first alternative to GPT; makes different trade-offs.

Example

Claude 3.5 Sonnet (2024) matches GPT-4 on many benchmarks while showing fewer false refusals and better instruction-following.

How it actually works

Claude uses constitutional AI instead of just RLHF; models are trained to be helpful but also to point out problems in requests. Accessible via API and web chat. Opus is the largest/most capable; Sonnet is the balanced choice; Haiku is fast.

For product teams

Differentiation: alignment is a feature, not a liability. Attracts safety-conscious enterprises.

For engineers

Constitutional AI leads to fewer false refusals and more nuanced behavior. Long context (200k tokens) is a practical advantage.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome