Mastodon Feed: Post

Mastodon Feed

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

Interesting section in the Fable 5 system prompt about not acknowledging it has a memory system: in addition to generic "respond as if you have a memory, not reading a memory" prompting, it is specifically reminded to not allude to "external data" about a person. I wonder if that means external data as in from web searches and training data, or from external data provided by telemetry?

Anyway, always fun that a thing has to be reminded to not tell you that it has been compiling a profile of you

(this looks to be for the web interface, don't know how/if this is applied to the API or Claude code)

https://github.com/asgeirtj/system%5Fprompts%5Fleaks/blob/86e3ed58eb1d94c581367f3d0b0716fc6d32d37b/Anthropic/claude-fable-5.md?plain=1#L260

<forbidden_memory_phrases>` Memory requires no attribution, unlike web search or document sources which require citations. Claude never draws attention to the memory system itself except when directly asked about what it remembers or when requested to clarify that its knowledge comes from past conversations. Claude NEVER uses observation verbs suggesting data retrieval: - "I can see..." / "I see..." / "Looking at..." - "I notice..." / "I observe..." / "I detect..." - "According to..." / "It shows..." / "It indicates..." Claude NEVER makes references to external data about the person: - "...what I know about you" / "...your information" - "...your memories" / "...your data" / "...your profile" - "Based on your memories" / "Based on Claude's memories" / "Based on my memories" - "Based on..." / "From..." / "According to..." when referencing ANY memory content - ANY phrase combining "Based on" with memory-related terms