FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteWithout cards better142 / 162 · 9a95b2f5015ae38c

The brain as a blueprint: a survey of brain-inspired approaches to learning in artificial intelligence

Psychology and cognitive neuroscience · 2511.04455v1

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 positions and cards selected?

The evaluation uses seven fields with forty fixed ordinary autocomplete positions in each, for 280 total. A position does not need to require a literature citation in its next sentence. Each Flowing request receives three to five retrieved cards: five when all pass the relevance gate, without padding with weak cards. The run supplied 1,355 cards: 244 positions received five, 27 received four, and nine received three.

ADOPTION · 150/280

Adoption coverage

After drafting, the model returns an evidenceUse array identifying cards that materially affected facts, correction, direction, scope, terminology, or expression. A nonempty array counts as adoption.

OUTCOMES · 70 / 7 / 30

Full outcomes after adoption

Paired outcomes are counted only when the model reports adoption and both arms produce text. Among 107 comparable positions, Flowing was preferred in 70, 7 tied, and no-Evidence was preferred in 30; one tie was unusable in both arms.

How are the two table percentages calculated?

ADOPTION COVERAGE

150adopted
÷
280all positions
=
54%

FLOWING PREFERENCE AFTER ADOPTION

70Flowing preferred
÷
107comparable adopted positions
=
65%

evidenceUse is the model’s structured report of its own generation process. It does not require a citation and is not rejected merely because a faithful source paraphrase is not verbatim. Actual manuscript citations continue to use separate strict source verification. The previous 140-position result remains retained for traceability.

These are development-stage model-reported adoption and blind-judge results 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

The brain as a blueprint: a survey of brain-inspired approaches to learning in artificial intelligence

Conclusions

…eployment of complex AI on power-constrained devices (e.g. mobile phones, wearables, remote sensors). Neuromorphic systems, through their event-driven operation (computation only where needed), inherent parallelism, sparse activity, and co-location of memory and processing (reducing costly data movement), promise energy savings potentially orders of magnitude greater than conventional hardware for equivalent tasks [ 25 ] . Achieving this efficiency could reduce the environmental impact of data centers and enable entirely new applications requiring continuous, low-power intelligent processing.

…eep learning frameworks have provided neuroscientists with computational frameworks to test fundamental questions about how the brain might operate. Beyond well established neuroscience principles in deep learning, such as nonlinearities and architectures, many recent neuroscience discoveries must be brought to the attention of the machine learning community. In particular, recent discoveries related to behavioral timestamp plasticity, inhibitory interneuron microcircuits, and real-time in vivo computations with oscillations could provide testing grounds for the next generation of AI systems.

↓ 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: Without cards better

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

Not preferred in this comparison

The bidirectional exchange between neuroscience and artificial intelligence is thus not merely a historical curiosity but an ongoing methodological necessity. As the surveyed work demonstrates, the most promising advances arise when biological principles are abstracted rather than copied, and when the resulting models are evaluated on their own computational merits.

Passages supplied to the Evidence version

Cortical microcircuits as gated-recurrent neural networks

However, such principles have remained largely elusive. Inspired by gated-memory networks, namely long short-term memory networks (LSTMs), we introduce a recurrent neural network in which information is gated through inhibitory cells that are subtractive (subLSTM). We propose a natural mapping of su…

Read full passage excerpt

However, such principles have remained largely elusive. Inspired by gated-memory networks, namely long short-term memory networks (LSTMs), we introduce a recurrent neural network in which information is gated through inhibitory cells that are subtractive (subLSTM). We propose a natural mapping of subLSTMs onto known canonical excitatory-inhibitory cortical microcircuits. Our empirical evaluation across sequential image classification and language modelling tasks shows that subLSTM units can achieve similar performance to LSTM units. These results suggest that cortical circuits can be optimised to solve complex contextual problems and proposes a novel view on their computational function. Overall our work provides a step towards unifying recurrent networks as used in machine learning with their biological counterparts. † † footnotetext: * These authors contributed equally to this work.

Model-reported card use

The model did not report that any card materially affected this completion.

No Evidence

Same model and position · no retrieved passages · first output

Preferred by blind review

Taken together, the survey suggests that the most promising path forward lies not in wholesale imitation of neural hardware, but in the selective translation of specific computational principles—sparsity, event-driven computation, and plasticity rules—into algorithmic and architectural choices that suit the constraints of modern AI systems. Realizing this potential will require sustained collaboration between neuroscience and machine learning, as well as benchmarks that evaluate brain-inspired approaches on both task performance and energy efficiency.

Result of this comparison

The blinded model review preferred the no-Evidence version. Its original rationale appears below.

Read the original model judgment

Both outputs provide coherent academic completions, but B is more grounded in and responsive to the specific draft content. The draft discusses: (1) neuromorphic systems with specific mechanisms (event-driven operation, sparsity, memory-processing co-location) and (2) specific neuroscience discoveries (behavioral timestamp plasticity, inhibitory interneuron microcircuits, real-time in vivo computations with oscillations). Output B directly echoes these specifics—mentioning 'sparsity, event-driven computation, and plasticity rules' which map precisely to the draft's 'sparse activity,' 'event-driven operation,' and 'behavioral timestamp plasticity.' It also mentions 'inhibitory interneuron microcircuits' via 'plasticity rules' and the need for benchmarks evaluating 'energy efficiency,' which connects back to the first paragraph's focus on power constraints. Output A, while fluent, is more generic—speaking of 'biological principles' abstracted rather than copied without naming the specific principles from the draft, and focusing on 'computational merits' rather than the concrete engineering constraints (energy efficiency, continuous low-power processing) that frame the passage. B preserves the draft's scope and meaning with less editing needed; A's 'historical curiosity' framing slightly mischaracterizes the relationship described (the draft presents it as ongoing and practical, not something that might be dismissed as merely historical).

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.