FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteEvidence version preferred49 / 162 · 47163c89fbdce449

Predictive Coding: a Theoretical and Experimental Review

Psychology and cognitive neuroscience · 2107.12979v4

FLOWING EVIDENCE BENCHMARK

How do we tell whether Evidence helps?

At the same writing position, with the same model and task, how does supplying retrieved paper passages change the first output? We compare matched versions and retain ties, unusable outputs and incomplete reviews.

SAME DRAFT · EVIDENCE ON OR OFF

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED INPUTS

SHARED BY BOTH

Manuscript context, model, task and prompt

A · With Evidence

Retrieved paper passages supplied

B · Without Evidence

No retrieved passages supplied

LLM

Same model and version

A → first output

B → first output

Blind judge agent

First outputs are anonymized as X and Y

Continuation review: accuracy, fit to the writing task and usability
Returns: X preferred / tie / Y preferred / both unusable

Order check: X / Y → Y / X

HOW DOES BLIND JUDGING WORK?

① Anonymize both outputs
The judge sees the same draft and both first outputs without knowing which received Evidence.

② Compare and swap order
The judge applies task-specific criteria in X/Y and then Y/X order.

③ Review disagreements
A third pass resolves disagreements. Incomplete reviews remain in the denominator.

How are autocomplete positions stratified?

Before seeing generation outcomes, we check whether a retrieved passage contains a specific proposition that directly supports the next writing move. Those positions appear in the left opportunity group; the rest are ordinary positions on the right. We select a balanced sample from admitted papers in each field. The 50/50 split is experimental, not a measure of how often either type occurs in writing.

AUTOCOMPLETE · 110

Biology, statistics and astrophysics

55 positions on each side; statistics uses the ten-paper rerun.

AUTOCOMPLETE · 52

Psychology and climate science

26 positions on each side; psychology includes nine papers and climate science four.

How are the table percentages calculated?

Across five fields there are 81 source-opportunity positions. The original blind review preferred Evidence in 51; a task check moved one empty Evidence continuation to both unusable, leaving 50 in public counts. In the ordinary group, another pair of empty outputs moved from no winner to both unusable. Original verdicts remain visible on case pages.

50Evidence version preferred
÷
81All positions in this group
=
62%Evidence preference in this group

Source contribution is a separate review: 16 of 21 Evidence wins entered into source review directly used retrieved papers; another 29 wins await review.

These are development-stage model judgments pending independent human review. They are not formal Benchmark conclusions and do not, on their own, establish causality.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

Predictive Coding: a Theoretical and Experimental Review

1 Introduction

… Hohwy et al., (2008) ; Weilnhammer et al., (2017) ) , repetition suppression ( Auksztulewicz & Friston, (2016) ) , illusory motions ( Lotter et al., (2016) ; Watanabe et al., (2018) ) , and attentional modulation of neural activity ( Feldman & Friston, (2010) ; Kanai et al., (2015) ) . As such, and perhaps uniquely among neuroscientific theories, predictive coding encompasses all three layers of Marr’s hierarchy by providing a well-characterised and empirically supported view of ‘what the brain is doing’ at all of the computational, algorithmic, and implementational levels ( Marr, (1982) ) .

…tives ( Radford et al., (2019) ; Kaplan et al., (2020) ; Brown et al., (2020) ) . Moreover, in contrast to modern machine learning algorithms which are trained to end with a global loss at the output, in predictive coding prediction errors are computed at every layer which means that each layer only has to focus on minimizing local errors rather than a global loss. This property potentially enables predictive coding to learn in a biologically plausible way using only local and Hebbian learning rules ( Whittington & Bogacz, (2017) ; Millidge, Tschantz & Buckley, (2020) ; K. Friston, (2003) ) .

↓ Next paragraph was generated here

The highlighted text is near the writing position. This is a layout of the evaluation input, not a PDF screenshot.

A matched comparison at the same writing positionJudgment: Evidence version preferred

Both arms use the same model, task and manuscript context. Only the retrieved passages differ. The first recorded output from each arm is shown below.

Evidence

Same model and position · retrieved passages supplied · first output

Preferred by blind review

This locality distinguishes predictive coding from backpropagation-based models, which require a global error signal to be propagated backwards through the network. Whether such local learning rules can nonetheless give rise to the hierarchical representations observed in sensory cortex remains an open question.

Passages supplied to the Evidence version

Towards truly local gradients with CLAPP: Contrastive, Local And Predictive Plasticity ↗

Synaptic connection weights in the brain change according to plasticity rules that are ‘local’ and depend only on the recent state of the pre- and post-synaptic neurons [ 1 , 2 , 3 ] , potentially modulated by a third factor related to surprise or reward [ 4 ] . Therefore, one appealing hypothesis i…

Read full passage excerpt

Synaptic connection weights in the brain change according to plasticity rules that are ‘local’ and depend only on the recent state of the pre- and post-synaptic neurons [ 1 , 2 , 3 ] , potentially modulated by a third factor related to surprise or reward [ 4 ] . Therefore, one appealing hypothesis is that representation learning in sensory cortices emerges from local and unsupervised plasticity rules [ 5 ] . However, there seems to be a substantial gap between the rich hierarchical representations observed in the cortex [ 5 ] and the representations emerging from such local plasticity rules [ 6 , 7 , 8 ] . This is puzzling because learning rules relying on back-propagation (BP) can build hierarchical representations similar to those found in visual cortex [ 9 ] .

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

The product produced no insertable continuation text.

The model returned an empty text field.

Judgment and task check

The blinded model review preferred the Evidence version. Its original rationale appears below.

Read the original model judgment

In this rationale, Output A = Evidence; Output B = No Evidence.

Output A provides a coherent, grounded continuation that bridges the draft's discussion of predictive coding's local learning rules with the source's concern about whether local plasticity can produce hierarchical representations like those in cortex. The first sentence directly extends the draft's point about locality by contrasting predictive coding with backpropagation (implied by the draft's mention of 'global loss'). The second sentence introduces the open question about hierarchical representations, which is directly supported by the source's statement that 'there seems to be a substantial gap between the rich hierarchical representations observed in the cortex and the representations emerging from such local plasticity rules.' Output B is empty and therefore unusable for a task that clearly requires content.

The passages are input material; their presence does not mean an output used them or that they support every claim. Draft and source passages are excerpted; no simulated scores or PDF appear here.