Existing intervention libraries are often project-based, lack extensibility, are hard to maintain and share, and are limited to single or non-nested interventions on Transformers

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• The intervention is the basic primitive of pyvene, specified with a dict-based format rather than expressed as code executed at runtimeclaim0.778
  Design philosophy claim distinguishing pyvene's approach from prior libraries
• It is precisely those innovations which attempt to change the system of processes most deeply, that are hardest for society to accept, and therefore hardest to implement.claim0.771
• A vanilla hook-based approach, as in all previous libraries, fails to intervene on any recurrent or state-space modelclaim0.758
  Technical claim justifying pyvene's state-variable hook tracking for recurrent model support
• Software implementations for all of the models/behaviours presented are common for n = 2, and can be made very efficient for α_i that map many objects onto a much smaller set of object families.claim0.745
  Claim about current practical feasibility and efficiency of 2-way associative implementations.
• Many practical mechanistic projects can be satisfied by collecting sufficiently large intervention evaluation datasetsclaim0.743
  Optimistic practical note about mitigating divergence concerns without solving the theoretical problem
• Concept interventions on some concepts act as 'wrecking-ball' interventions, collapsing global model performance.finding0.739
  Observation of catastrophic performance drop when steering certain concepts.
• If the snippet works well, it may be adopted and spread to new construction methods, even in the context of different attitudes.hypothesis0.738
  Conditional prediction about the self-propagation of small effective sequences.
• Certain kinds of information structures actively facilitate their own transformation and remapping, exhibiting minimal agency.hypothesis0.738
  Speculative hypothesis that memories themselves are agents.