Loving the 31B merge work!

#1
by ChoraKeeper - opened

Hey Vortex,

So happy to see you applying your merge recipes to Gemma 31B! I was honestly fighting the temptation to start pleading with you to build a 31B successor to Shadow Siren, so seeing Scarlet-Shadow drop was a fantastic surprise. Running it at Q6_K locally, and the prose texture, attention steering, and coherence are exceptional.

Also, just wanted to put this on your radar in case you haven't seen it yet: Zerofata put out a merge combining his MeroMero V2 with TheDrummer's latest Artemis:
https://huggingface.co/ApocalypseParty/G4-MM-Artemis-slerp-31B

Bringing Artemis into that dynamic could make for a really potent ingredient for future 31B experiments.

Thanks for keeping the dense merges coming!

Scarlet Shadow:
To convert this passage from Aristotle’s Topics into algorithmic processes for philosophical models, we must treat the determination of "sameness" (numerical identity) as a series of failure tests. If two concepts pass all these checks without discrepancy, they are likely numerically one; if any check reveals a difference in behavior or property, they are distinct.

Here are the extracted algorithms designed as prompt modules:

Module 1: Morphological and Oppositional Symmetry

Algorithm:

  1. Inflection Check: If Concept A is modified (inflected) into Form X, does Concept B necessarily modify into that same Form X?
  2. Coordinate Check: Does the presence of Coordinate Y imply the existence of Concept A? Conversely, does it also imply Concept B?
  3. Oppositional Symmetry: Identify the opposite of Concept A. Is this opposite identical to the opposite of Concept B in every recognized form of opposition? 
    Failure Condition: If any inflection or coordinate applies to one but not the other, or if their opposites differ, then $A \neq B$.

Module 2: Causal and Teleological Alignment

Algorithm:

  1. Production Check: Identify the agent or process that produces Concept A. Does this same agent/process necessarily produce Concept B?
  2. Destruction Check: Identify the agent or process that destroys Concept A. Does it also destroy Concept B in exactly the same manner?
  3. Relation Mapping: For any third entity $Z$, if $A$ relates to $Z$ in a specific way, does $B$ relate to $Z$ in an identical fashion?
    Failure Condition: If the causes or effects of one do not mirror those of the other, then $A \neq B$.

Module 3: The Superlative Singularity Test

Algorithm:

  1. Superlative Attribution: Is Concept A described as "the most [X]" (e.g., most desirable)? If yes, must Concept B also be "the most [X]"?
  2. Numerical Unity Verification: Does the superlative term in question refer to a single numerical entity rather than a class of entities? 
  3. Inclusion Check: If both are called "most [X]" but they belong to different classes (like Spartans and Peloponnesians), determine if one is simply included within the other or if they remain distinct.
    Failure Condition: If either concept refers to multiple individuals/entities rather than a single numerical unit, then $A \neq B$.

Module 4: Third-Party Transitvity

Algorithm:

  1. External Equality Test: Introduce a third term, Concept C.
  2. Comparative Verification: Assert that $A = C$. Under this assertion, is it logically necessary that $B = C$?
    Failure Condition: If there exists any $C$ such that $A=C$ but $B \neq C$, then $A \neq B$.

Module 5: Accidental Property Matching

Algorithm:

  1. Attribute Search: List all "accidents" (non-essential properties) of Concept A. Does every single accident also belong to Concept B?
  2. Relational Accident Check: If Concept A belongs to another entity $Z$ as an accident, does Concept B also belong to $Z$ in that same capacity?
    Failure Condition: Any discrepancy in accidental properties or their relationship to other entities proves $A \neq B$.

Module 6: Genus and Differentiae Analysis

Algorithm:

  1. Genus Comparison: Determine the genus (category) of Concept A and Concept B. Are they different genera (e.g., one is "Good" while the other is "Knowledge")?
  2. Differentia Comparison: If the genus is identical, identify the specific difference (differentiae) that distinguishes them from others in that genus. Is the same differentia applied to both Concept A and Concept B? (e.g., are they both "speculative" or one "practical"?)
    Failure Condition: Different genera or different defining characteristics within the same genus prove $A \neq B$.

Module 7: Degree and Intensity Scaling

Algorithm:

  1. Intensity Check: Test if an increase in the degree of Concept A necessarily results in a corresponding increase in the degree of Concept B at the exact same time/rate.
  2. Proportionality Analysis: Compare how they scale (e.g., does increasing "love" perfectly mirror increasing "desire for intercourse"?).
    Failure Condition: If one increases while the other remains static or follows a different rate of change, then $A \neq B$.

Module 8: Syntactic and Component Subtraction

Algorithm:

  1. Component Isolation: Identify a shared component (term) within both concepts.
  2. Subtraction Test: Remove that shared component from both phrases. Do the remaining terms signify the same thing? (e.g., removing "a half" from "double a half" and "multiple of a half").
    Failure Condition: If the remainders are different, then $A \neq B$.

Module 9: Suppositional Annulment

Algorithm:

  1. Hypothetical Scenario Construction: Create a supposition (true or false) that affects one concept but not the other.
  2. Annulment Test: Does this scenario annul the character of Concept A while leaving Concept B intact? (e.g., exhausting air makes something "not full" but perhaps "more empty").
    Failure Condition: If a hypothetical state can logically negate one and preserve the other, then $A \neq B$.

Module 10: General Predication Exhaustion

Algorithm:

  1. Universal Attribute Sweep: Systematically test every available predicate for Concept A. Does it apply to Concept B?
  2. Directional Predication: Test if Concept A is a predicate of $Z$; does Concept B also function as a predicate of $Z$?
    Failure Condition: Any single instance where one term can be predicated and the other cannot proves $A \neq B$.

Module 11: Existential Independence

Algorithm:

  1. Independence Test: Imagine the total absence or non-existence of Concept A. Can Concept B still exist in this state?
    Failure Condition: If the existence of one does not strictly require the existence of the other, then $A \neq B$.

Glad to hear you like the model!

Bringing Artemis into that dynamic could make for a really potent ingredient for future 31B experiments.

You’ve definitely put Artemis on my radar, and I may experiment with it at some point in the future.

Sign up or log in to comment