concept
active
concept:multitask-scaling

Multitask Scaling

The pressure on models trained on more tasks to find representations that generalize across all tasks, reducing the solution space

Neighborhood — ranked by edge-count

Concepts (1)

concept
  • The central empirical phenomenon: different neural networks trained on different data/objectives develop increasingly similar representations

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

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

  • Argues that there are fewer representations competent for N tasks than M<N tasks, so more general models have a smaller solution space
  • Used in the color cooccurrence experiment to embed colors into 3D space preserving dissimilarity matrix distances
  • Multi-scale agencyconcept0.776
    The nesting of overlapping, competing, and cooperating agents at different levels of organization, each with its own goals and cognitive boundary.
  • Systems comprising nested hierarchies (cells→organisms→societies); central to understanding collective intelligence
  • Hierarchical goal-seeking at multiple levels of organization.
  • Test-Time Scalingconcept0.771
    Approach using extra compute at test time to double-check answers and improve reliability.
  • Scaling Supervisionframework0.769
    Techniques that leverage AI to help humans more efficiently supervise AI.
  • Learning paradigm that jointly learns multiple related tasks using a single model