Capacity Hypothesis

Bigger models are more likely to converge to a shared representation than smaller models because they can better approximate the global optimum

Source paper

extracted_from

(2024) · Minyoung Huh · Brian Cheung · Tongzhou Wang · Phillip Isola

paper

finding

concept

Platonic Representation
supports
The hypothesized converged representation that all sufficiently large AI models are approaching — a statistical model of underlying reality

hypothesis

Multitask Scaling Hypothesis
associated_with
Argues that there are fewer representations competent for N tasks than M<N tasks, so more general models have a smaller solution space
Simplicity Bias Hypothesis
associated_with
Deep networks are biased toward finding simple fits to data, and this bias increases with model size, driving convergence

quote

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.

Reward Hypothesisconcept0.786
The claim in RL that any goal can be expressed as maximizing the expected cumulative sum of a scalar reward signal.
Theory of Mind Capacityconcept0.762
The cognitive ability to attribute mental states to self and others; used as a proxy domain for investigating consciousness in LLMs.
Genesis Hypothesisframework0.760
The conjecture that consciousness does not result from the organized mind but creates and maintains complex models of reality; forms at the beginning of mental development
Universality Hypothesisconcept0.756
The hypothesis that analogous features and circuits reliably form across different neural network models and tasks
Bayesian Brain Hypothesisframework0.742
Normative theory proposing biological systems perform approximate Bayesian inference through free energy minimization.
Machine Consciousness Hypothesisframework0.741
CIMC's central hypothesis: general computational machines with sufficient resources possess the necessary and sufficient means to implement consciousness, verifiable through internal structure analysis
I-hypothesisframework0.740
The overarching hypothesis that an I or self-like ground underlies matter and becomes visible in living things.
Privileged Bases Hypothesisconcept0.739
Hypothesis that neurons form privileged bases to encode information; consistent with constructive abstraction