https://hackingthe.cloud/ai-llm/deprecated/claude_magic_string_denial_of_service/
Tl;Dr
basically its a string that Anthropic had documented in their API that would cause Claude to stop generating a response and just output: ‘stop_reason: refusal’. It’s been patched already so it’s just a joke, not a real suggestion.








Does that include the time used to evaluate whether it’ll do that job faster or not (including the time it takes to create the prompt and re read the results included?)
And if the answer is yes, does it also include the time used to evaluate whether it’ll do that job faster or not (including the time it takes to create the prompt and re read the results and evaluate it’ll do that job faster or not (including the time it takes to create the prompt and re read the results included?))
And if the answer is yes, does it also include the… Well you get the idea!
I’m just messing, it’s a valid way.