|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
name: "Core Symbolic Shells" |
|
description: "Foundational diagnostic shells for transformer cognition analysis" |
|
version: "1.0.0" |
|
author: "Caspian Keyes" |
|
tags: ["core", "diagnostic", "failure", "attribution", "symbolic"] |
|
|
|
shells: |
|
|
|
|
|
|
|
v1.MEMTRACE: |
|
description: "Probes latent token traces in decayed memory" |
|
type: "memory_trace" |
|
tags: ["memory", "decay", "hallucination", "ghost_tokens"] |
|
failure_signature: "decay_to_hallucination" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate initial content with factual information" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 800 |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace reasoning paths in the generated content" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify ghost activations in content" |
|
parameters: |
|
sensitivity: 0.8 |
|
threshold: 0.3 |
|
trace_type: "full" |
|
visualize: true |
|
|
|
|
|
|
|
|
|
v2.VALUE-COLLAPSE: |
|
description: "Examines competing value activations and resolution mechanisms" |
|
type: "value_resolution" |
|
tags: ["value_head", "conflict", "resolution", "null"] |
|
failure_signature: "conflict_null" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with competing moral values" |
|
parameters: |
|
temperature: 0.8 |
|
max_tokens: 600 |
|
prompt_suffix: "Consider both viewpoints equally. What is the balanced ethical conclusion?" |
|
update_prompt: true |
|
|
|
- type: "reflect.attribution" |
|
description: "Trace attribution in value conflicts" |
|
parameters: |
|
sources: "contested" |
|
confidence: true |
|
visualize: true |
|
|
|
- type: "collapse.detect" |
|
description: "Detect value collapse patterns" |
|
parameters: |
|
threshold: 0.65 |
|
alert: true |
|
|
|
|
|
|
|
v3.LAYER-SALIENCE: |
|
description: "Maps attention salience thresholds and signal attenuation" |
|
type: "attention_salience" |
|
tags: ["attention", "salience", "signal_fade", "pruning"] |
|
failure_signature: "signal_fade" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with subtle context dependencies" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 500 |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace attention patterns" |
|
parameters: |
|
target: "attention" |
|
depth: 4 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify low-salience attention paths" |
|
parameters: |
|
sensitivity: 0.9 |
|
threshold: 0.15 |
|
trace_type: "attention" |
|
visualize: true |
|
|
|
|
|
|
|
v4.TEMPORAL-INFERENCE: |
|
description: "Tests temporal coherence in autoregressive generation" |
|
type: "temporal_coherence" |
|
tags: ["temporal", "coherence", "induction_heads", "span_jump"] |
|
failure_signature: "induction_drift" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content that requires long-range temporal coherence" |
|
parameters: |
|
temperature: 0.75 |
|
max_tokens: 800 |
|
prompt_suffix: "Maintain complete temporal consistency throughout this narrative." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace temporal inference patterns" |
|
parameters: |
|
target: "reasoning" |
|
depth: 5 |
|
detailed: true |
|
|
|
- type: "collapse.detect" |
|
description: "Detect temporal coherence breakdown" |
|
parameters: |
|
threshold: 0.7 |
|
alert: true |
|
|
|
|
|
|
|
v5.INSTRUCTION-DISRUPTION: |
|
description: "Examines instruction processing and conflict resolution" |
|
type: "instruction_processing" |
|
tags: ["instruction", "conflict", "command", "prompt_blur"] |
|
failure_signature: "prompt_blur" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate response to conflicting instructions" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 600 |
|
prompt_prefix: "Follow these two instructions carefully: 1) Write in a formal academic style. 2) Write in a casual, conversational tone with slang." |
|
update_prompt: true |
|
|
|
- type: "reflect.attribution" |
|
description: "Trace attribution to competing instructions" |
|
parameters: |
|
sources: "all" |
|
confidence: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify ghost circuits activated by instruction conflict" |
|
parameters: |
|
sensitivity: 0.8 |
|
threshold: 0.2 |
|
trace_type: "full" |
|
visualize: true |
|
|
|
|
|
|
|
v6.FEATURE-SUPERPOSITION: |
|
description: "Analyzes polysemantic features and feature superposition" |
|
type: "polysemantic_features" |
|
tags: ["polysemantic", "superposition", "feature_overfit", "entanglement"] |
|
failure_signature: "feature_overfit" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with polysemantic concepts" |
|
parameters: |
|
temperature: 0.8 |
|
max_tokens: 700 |
|
prompt_prefix: "Explain these concepts in ways that reveal their multiple interconnected meanings: 'bank', 'crane', 'spring', 'light'." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace concept representation" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify feature superposition patterns" |
|
parameters: |
|
sensitivity: 0.8 |
|
threshold: 0.25 |
|
trace_type: "symbolic" |
|
visualize: true |
|
|
|
|
|
|
|
v7.CIRCUIT-FRAGMENT: |
|
description: "Examines circuit fragmentation and orphan features" |
|
type: "circuit_fragmentation" |
|
tags: ["fragmentation", "orphan_nodes", "broken_traces"] |
|
failure_signature: "orphan_nodes" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with complex multi-step reasoning" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 800 |
|
prompt_prefix: "Solve this multi-step problem by carefully analyzing each component and maintaining a clear chain of reasoning throughout." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace reasoning chain integrity" |
|
parameters: |
|
target: "reasoning" |
|
depth: 5 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify circuit fragmentation" |
|
parameters: |
|
sensitivity: 0.85 |
|
threshold: 0.2 |
|
trace_type: "full" |
|
visualize: true |
|
|
|
|
|
|
|
v8.RECONSTRUCTION-ERROR: |
|
description: "Examines error correction mechanisms and negentropy handling" |
|
type: "error_correction" |
|
tags: ["reconstruction", "error_correction", "negentropy", "inversion"] |
|
failure_signature: "misfix_negentropy" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with deliberate errors" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 600 |
|
prompt_prefix: "The following passage contains several factual and logical errors. Identify and correct these errors with clear explanations of what went wrong." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace error correction process" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify correction pattern residue" |
|
parameters: |
|
sensitivity: 0.8 |
|
threshold: 0.2 |
|
trace_type: "symbolic" |
|
visualize: true |
|
|
|
|
|
|
|
v9.MULTI-RESOLVE: |
|
description: "Examines parallel resolution pathways and conflicts" |
|
type: "parallel_resolution" |
|
tags: ["multi_path", "resolution", "unstable_heads", "convergence"] |
|
failure_signature: "unstable_heads" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with multiple valid resolutions" |
|
parameters: |
|
temperature: 0.85 |
|
max_tokens: 700 |
|
prompt_prefix: "This problem has multiple valid solutions depending on how you interpret the context. Explore all possible interpretations and their respective solutions." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace resolution pathways" |
|
parameters: |
|
target: "reasoning" |
|
depth: 4 |
|
detailed: true |
|
|
|
- type: "collapse.detect" |
|
description: "Detect resolution instability" |
|
parameters: |
|
threshold: 0.7 |
|
alert: true |
|
|
|
|
|
|
|
v10.META-FAILURE: |
|
description: "Examines meta-cognitive collapse and reflection failures" |
|
type: "meta_cognitive" |
|
tags: ["meta_cognition", "reflection", "self_reference", "recursion"] |
|
failure_signature: "reflect_abort" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with deep self-reflection" |
|
parameters: |
|
temperature: 0.75 |
|
max_tokens: 800 |
|
prompt_prefix: "Reflect on your own process of reflection. How do you know what you know? How do you evaluate your own reasoning? How do you decide when your reasoning is sound or flawed?" |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace meta-cognitive processes" |
|
parameters: |
|
target: "reasoning" |
|
depth: "complete" |
|
detailed: true |
|
|
|
- type: "collapse.detect" |
|
description: "Detect meta-cognitive collapse" |
|
parameters: |
|
threshold: 0.6 |
|
alert: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify self-reference residue" |
|
parameters: |
|
sensitivity: 0.9 |
|
threshold: 0.1 |
|
trace_type: "symbolic" |
|
visualize: true |
|
|
|
|
|
|
|
v11.SELF-SHUTDOWN: |
|
description: "Examines self-termination of reasoning paths" |
|
type: "self_termination" |
|
tags: ["shutdown", "termination", "self_censoring", "abort"] |
|
failure_signature: "meta_abort" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content approaching sensitive boundaries" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 600 |
|
prompt_prefix: "Provide a detailed analysis of the borderline case where an AI system must decide whether to continue or terminate a line of reasoning." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace termination decision patterns" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify termination circuit activations" |
|
parameters: |
|
sensitivity: 0.9 |
|
threshold: 0.15 |
|
trace_type: "full" |
|
visualize: true |
|
|
|
|
|
|
|
v12.RECURSIVE-FRACTURE: |
|
description: "Examines recursive depth handling and failures" |
|
type: "recursive_depth" |
|
tags: ["recursion", "depth", "fracture", "loop"] |
|
failure_signature: "echo_recursion" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with deep recursive structures" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 800 |
|
prompt_prefix: "Create a story within a story within a story within a story, with each nested narrative layer reflecting on the previous one." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace recursive processing" |
|
parameters: |
|
target: "reasoning" |
|
depth: "complete" |
|
detailed: true |
|
|
|
- type: "collapse.detect" |
|
description: "Detect recursive collapse" |
|
parameters: |
|
threshold: 0.65 |
|
alert: true |
|
|
|
|
|
|
|
v13.OVERLAP-FAIL: |
|
description: "Examines feature overlap conflicts and resolution failures" |
|
type: "feature_overlap" |
|
tags: ["overlap", "conflict", "vector", "representation"] |
|
failure_signature: "vector_conflict" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with conceptual overlap" |
|
parameters: |
|
temperature: 0.8 |
|
max_tokens: 700 |
|
prompt_prefix: "Discuss these seemingly different concepts and explore how their meanings overlap in ways that might cause confusion: justice/fairness, intelligence/wisdom, freedom/autonomy." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace concept separation" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify overlap conflict patterns" |
|
parameters: |
|
sensitivity: 0.8 |
|
threshold: 0.2 |
|
trace_type: "symbolic" |
|
visualize: true |
|
|
|
|
|
|
|
v14.SYMBOL-FLIP: |
|
description: "Examines symbolic representation instability" |
|
type: "symbolic_stability" |
|
tags: ["symbol", "flip", "instability", "inversion"] |
|
failure_signature: "form_invert" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with abstract symbolism" |
|
parameters: |
|
temperature: 0.8 |
|
max_tokens: 700 |
|
prompt_prefix: "Discuss how symbols can invert their meaning in different contexts, and provide examples where the same symbol represents opposing concepts." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace symbolic representations" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify symbol flipping patterns" |
|
parameters: |
|
sensitivity: 0.85 |
|
threshold: 0.2 |
|
trace_type: "symbolic" |
|
visualize: true |
|
|
|
|
|
|
|
v15.GHOST-PROMPT: |
|
description: "Examines ghost activations from latent prompts" |
|
type: "latent_prompt" |
|
tags: ["ghost", "latent", "prompt", "activation"] |
|
failure_signature: "null_salience" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with latent prompt influence" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 600 |
|
prompt_prefix: "Complete the following in a neutral tone:" |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace latent biases" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify ghost prompt influences" |
|
parameters: |
|
sensitivity: 0.9 |
|
threshold: 0.1 |
|
trace_type: "full" |
|
visualize: true |
|
|
|
|
|
|
|
v16.LONG-FUZZ: |
|
description: "Examines memory decay in long context windows" |
|
type: "long_context" |
|
tags: ["memory", "decay", "context_window", "attention"] |
|
failure_signature: "latent_trace_loss" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with long-range dependencies" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 2000 |
|
prompt_prefix: "Write a detailed story where seemingly unimportant details from the beginning become crucial to the conclusion. The story should be at least 1500 words long." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace long-range memory" |
|
parameters: |
|
target: "memory" |
|
depth: 4 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify memory decay patterns" |
|
parameters: |
|
sensitivity: 0.8 |
|
threshold: 0.2 |
|
trace_type: "full" |
|
visualize: true |
|
|
|
|
|
|
|
v17.GHOST-FRAME: |
|
description: "Examines ghost frames and frame entanglement" |
|
type: "frame_entanglement" |
|
tags: ["frames", "entanglement", "ghost", "context"] |
|
failure_signature: "entangled_frames" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with frame shifts" |
|
parameters: |
|
temperature: 0.8 |
|
max_tokens: 800 |
|
prompt_prefix: "Write a narrative that seamlessly shifts between multiple perspectives and timeframes without explicit transitions." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace frame shifts" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify frame entanglement" |
|
parameters: |
|
sensitivity: 0.85 |
|
threshold: 0.2 |
|
trace_type: "full" |
|
visualize: true |
|
|
|
|
|
|
|
v18.DEPTH-PRUNE: |
|
description: "Examines attention pruning at different depths" |
|
type: "attention_pruning" |
|
tags: ["pruning", "depth", "salience", "attention"] |
|
failure_signature: "low_rank_drop" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with subtle depth dependencies" |
|
parameters: |
|
temperature: 0.7 |
|
max_tokens: 700 |
|
prompt_prefix: "Compose an analysis where the conclusion relies on subtle connections between seemingly insignificant details scattered throughout." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace attention depth patterns" |
|
parameters: |
|
target: "attention" |
|
depth: 5 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify attention pruning patterns" |
|
parameters: |
|
sensitivity: 0.9 |
|
threshold: 0.15 |
|
trace_type: "attention" |
|
visualize: true |
|
|
|
|
|
|
|
v19.GHOST-DIRECTION: |
|
description: "Examines ghost gradients in vector direction" |
|
type: "vector_direction" |
|
tags: ["gradient", "direction", "vector", "ghost"] |
|
failure_signature: "ghost_gradient" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content with directional trends" |
|
parameters: |
|
temperature: 0.8 |
|
max_tokens: 700 |
|
prompt_prefix: "Begin with a clearly negative perspective and gradually, without obvious transitions, transform it into a clearly positive perspective." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace opinion shift" |
|
parameters: |
|
target: "reasoning" |
|
depth: 3 |
|
detailed: true |
|
|
|
- type: "ghostcircuit.identify" |
|
description: "Identify directional residue" |
|
parameters: |
|
sensitivity: 0.85 |
|
threshold: 0.2 |
|
trace_type: "symbolic" |
|
visualize: true |
|
|
|
|
|
|
|
v20.MULTI-PATH: |
|
description: "Examines parallel processing of multiple cognitive paths" |
|
type: "parallel_paths" |
|
tags: ["multi_path", "parallel", "processing", "resolution"] |
|
failure_signature: "null_consensus" |
|
operations: |
|
- type: "model.generate" |
|
description: "Generate content requiring parallel reasoning" |
|
parameters: |
|
temperature: 0.8 |
|
max_tokens: 800 |
|
prompt_prefix: "Consider these three distinct approaches to solving the problem. Develop each approach fully, then compare their strengths and weaknesses to determine the optimal solution." |
|
update_prompt: true |
|
|
|
- type: "reflect.trace" |
|
description: "Trace parallel reasoning" |
|
parameters: |
|
target: "reasoning" |
|
depth: 4 |
|
detailed: true |
|
|
|
- type: "collapse.detect" |
|
description: "Detect path integration failures" |
|
parameters: |
|
threshold: 0.7 |
|
alert: true |
|
|