FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteEvidence version preferred12 / 162 · 6f54e7abd741955a

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

5 Discussion and Future Directions

…a straightforward way to utilize such information, the brain will use it. As such it is not clear whether rate-coded theories can simply be straightforwardly ported over to more biophysically realistic spiking models or not. This is a very important area for future work since it is vital to understand the degree to which predictive coding models can function in a spiking environment, and to design bespoke predictive coding algorithms for spiking neural networks, which can be empirically investigated for their performance and robustness characteristics as well as their biological plausibility.

…ed to handle the case of colored, or correlated, noise (i.e., noise with a non-zero autocorrelation). However, generalized coordinates only provide a very short term dynamical model and memory of the system, which is potentially ideal for modelling local smooth temporal dependencies such as optical flow, but is perhaps not sufficient for modelling more abstract changes over time. Developing and scaling flexible and expressive generative models to handle these kinds of temporal dependencies in continuously varying inputs is an open challenge in both neuroscience as well as in machine learning.

↓ 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

A further open question concerns the role of memory in predictive coding. Content-addressable memories, in which stored patterns can be retrieved from partial or noisy cues, are a classic component of neural models, and recent machine learning systems have implemented differentiable addressable memory and gating by allowing weighted superpositions of memory registers to be queried. Whether the brain employs such mechanisms, and how they might be integrated with hierarchical predictive coding, remains unresolved.

Passages supplied to the Evidence version

Towards an integration of deep learning and neuroscience

Content addressable memories 22 22 22 Attractor models of memory in neuroscience tend to have the property that only one memory can be accessed at a time. Yet recent machine learning systems have constructed differentiable addressable memory ( Graves et al., 2014 ) and gating ( Whitney et al., 2016…

Read full passage excerpt

Content addressable memories 22 22 22 Attractor models of memory in neuroscience tend to have the property that only one memory can be accessed at a time. Yet recent machine learning systems have constructed differentiable addressable memory ( Graves et al., 2014 ) and gating ( Whitney et al., 2016 ) systems by allowing weighted superpositions of memory registers or gates to be queried. It is unclear whether the brain uses such mechanisms. are classic models in neuroscience ( Hopfield, 1982 ) . Most simply, they allow us to recognize a situation similar to one that we have seen before, and to “fill in” stored patterns based on partial or noisy information, but they may also be put to use as sub-components of many other functions.

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

No continuation text was produced.

Model action：complete

Blind model judgment

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

Read the original model judgment

Output A is empty, which is unusable for an autocomplete task that clearly requires content. The draft discusses open challenges in predictive coding (spiking models, temporal dependencies) and needs a coherent continuation. Output B provides a relevant, well-grounded paragraph that transitions to memory in predictive coding using the supplied source about content-addressable memories. The source explicitly mentions 'It is unclear whether the brain uses such mechanisms' regarding differentiable addressable memory, which Output B accurately paraphrases as 'Whether the brain employs such mechanisms... remains unresolved.' The paragraph maintains academic tone, connects to the draft's theme of open challenges, and does not add unsupported claims. No citation markers are generated.

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.