FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteEvidence version preferred9 / 162 · 199cc69dbcc484e8

Predictive Coding: a Theoretical and Experimental Review

Quantitative biology · 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, 51 of 81 left-column positions preferred the Evidence version. Ties, pairs where both versions were unusable and incomplete reviews remain in the denominator.

51Evidence version preferred
÷
81All positions in this group
=
63%Evidence preference in this group

Source contribution is a separate review: 16 of 22 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

… we provide a comprehensive review both of the core mathematical structure and logic of predictive coding, thus complementing recent tutorials in the literature ( Buckley et al. (2017) ; Bogacz (2017) ) . We also review a wide range of classic and recent work within the framework, ranging from the neurobiologically realistic microcircuits that could implement predictive coding, to the close relationship between predictive coding and the widely-used backpropagation of error algorithm, as well as surveying the close relationships between predictive coding and modern machine learning techniques.

↓ 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

Within this literature, predictive coding has been developed as a biologically plausible alternative to backpropagation for credit assignment in arbitrary computation graphs, with reported performance comparable to backprop on architectures such as CNNs and LSTMs.

Passages supplied to the Evidence version

Applications of the Free Energy Principle to Machine Learning and Neuroscience

Specifically, in Chapter 6, we first provide a brief review of the credit assignment problem in the brain, as well as the backpropagation algorithm (and automatic differentiation in general), for context, and then present our two contributions to this field. First, we demonstrate how under certain c…

Read full passage excerpt

Specifically, in Chapter 6, we first provide a brief review of the credit assignment problem in the brain, as well as the backpropagation algorithm (and automatic differentiation in general), for context, and then present our two contributions to this field. First, we demonstrate how under certain conditions, predictive coding itself can be utilized as a biologically plausible method of credit assignment in the brain, can apply to any arbitrary computation graph, and can be used to train modern machine learning architectures such as CNNs and LSTMs with performance comparable to backprop.

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

Predictive coding has been developed independently in neuroscience and machine learning, where it is known as predictive coding or, in some formulations, as a variational inference scheme for hierarchical generative models. This dual lineage has produced a rich but sometimes fragmented literature, with related ideas appearing under different names in the two fields.

Blind model judgment

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

Read the original model judgment

Output B is tightly grounded in the supplied source, which explicitly states that predictive coding 'can be utilized as a biologically plausible method of credit assignment in the brain, can apply to any arbitrary computation graph, and can be used to train modern machine learning architectures such as CNNs and LSTMs with performance comparable to backprop.' Output B accurately reflects these claims with appropriate hedging ('reported performance'). Output A, however, contains multiple unsupported claims: (1) that predictive coding was 'developed independently in neuroscience and machine learning' — the source does not address developmental history or independence; (2) that it is 'known as predictive coding or, in some formulations, as a variational inference scheme for hierarchical generative models' — the source does not mention variational inference or hierarchical generative models; (3) that 'This dual lineage has produced a rich but sometimes fragmented literature, with related ideas appearing under different names in the two fields' — this is a broad historiographical claim with no support in the source. These unsupported factual claims in Output A make it less suitable for direct insertion despite its fluent prose.

Source review confirmed that this continuation directly used specific, verifiable information from the retrieved paper.

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.