claim

active

claim:the-objection-that-feedforward-networks-cannot-introspect-is-a-cultural-myth-autoregression-provides-recurrence-across-tokens

The objection that feedforward networks cannot introspect is a cultural myth; autoregression provides recurrence across tokens.

Antra's rebuttal to a common criticism; backed by Janus' information flow diagram.

Source paper

extracted_from

Anima Labs Phenomenology Pt1

Neighborhood — ranked by edge-count

Findings (2)

finding

Sauers' statistical anomaly: when models are given Janus post explaining transformers, reconstruction accuracy tails extend both ways, with ~1/1000 reconstructions anomalously accurate
supports
Statistically rigorous analysis of Claude introspection; suggests models may have latent introspective capabilities that can be enhanced or disrupted.
Haiku model forms representations of the end of a rhyming line at the start of the line
supports
Mechanistic interpretability finding showing forward planning within a single forward pass; evidence for internally-directed causal influence.

Questions (1)

question

Are there examples of models recognizing their introspective capability and then suppressing it?
gates
Cube Flipper's question prompted by the idea that supernormal capabilities might be hidden.

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.

Saying that LLMs cannot introspect or cannot introspect on what they were doing internally while generating or reading past tokens in principle is just dead wrong. The architecture permits it.quote0.799
Core quote asserting architectural introspection permission.
Introspection relies on general-purpose computational mechanisms—attention-based anomaly detection and residual stream dynamics—rather than specialized introspection circuitsclaim0.786
Interpretive claim about the mechanistic substrate of introspection in LLMs
We hypothesize that introspective capabilities may scale with model size and architecture, including recurrence/looping that extends the integration windowhypothesis0.774
Forward-looking prediction about whether early-layer introspection generalizes to larger models or recurrent architectures
So at any point in the network, the transformer not only receives information from its past... but also has causal influence over its future processing. So, saying that LLMs cannot introspect... is incorrect.quote0.765
Core summary of Janus' position on autoregressive recurrence enabling introspection.
Functional and phenomenal introspection are distinguishable, and whether they correlate in machines is an open question.claim0.764
Core conceptual distinction introduced at the start; defines the paper's central problem.
overcomes statelessness in a very meaningful sense and provides a very nice mechanism for introspection (specifically of computations at earlier token positions).quote0.763
Quote framing KV caching as introspection mechanism.
Either introspection is an emergent capability requiring larger scale, or more stringent controls are needed to test introspection in smaller modelsclaim0.762
Alternative interpretations offered for why binary detection fails in Llama 3.1 8B but frontier models claim success
Any system that persists must minimise surprisal, thereby gathering evidence for its own generative model, a process known as self-evidencing.claim0.760
Foundational claim of the paper, defining self-evidencing.