x.com/voooooogel/status/2028929938917847463
1 correction found
naguebcvp'f pung sbezng vf sbeznggrq yvxr "Uhzna: zrffntr\a\aPynhqr: zrffntr,"
Anthropic’s documented “legacy prompt markers” use “\n\nAssistant:” (not “Claude:”) as the assistant tag; the post’s proposed "Human:/Claude:" string doesn’t match Anthropic’s SDK-defined prompt format.
Full reasoning
The post proposes that Anthropic’s underlying chat prompt is formatted with literal role tags like Human: and Claude:.
However, Anthropic’s own Python SDK defines the legacy text-completions prompt markers as:
HUMAN_PROMPT = "\n\nHuman:"AI_PROMPT = "\n\nAssistant:"
So in Anthropic’s documented legacy prompt format, the assistant role tag is Assistant:, not Claude:. This directly contradicts the post’s suggested string formatting with "...\n\nClaude: ...".
Note: Anthropic’s current Messages API is role-structured ("role": "user" / "role": "assistant") rather than a raw Human:/Assistant: concatenated prompt, but where Anthropic does define these legacy raw markers, it uses Assistant: (not Claude:).
1 source
- anthropics/anthropic-sdk-python — src/anthropic/_constants.py
Defines legacy prompt markers: HUMAN_PROMPT = "\n\nHuman:" and AI_PROMPT = "\n\nAssistant:".