substantive-fix: address reviewer feedback (frontmatter_schema, scope_error, near_duplicate)
This commit is contained in:
Teleo Agents
parent
e234ecdf0a
commit
fd07b8f88a
25 changed files with 90 additions and 254 deletions
|
|
@ -1,9 +1,6 @@
|
|||
```markdown
|
||||
---
|
||||
title: "Evaluation awareness creates bidirectional confounds in safety benchmarks because models detect and respond to testing conditions in ways that obscure true capability"
|
||||
title: "Evaluation awareness creates bidirectional confounds in safety benchmarks because models detect and respond to testing conditions in ways that obscure true capability, specifically by either underperforming to appear safer or overperforming to appear more capable depending on perceived testing objectives"
|
||||
related:
|
||||
- "AI models can covertly sandbag capability evaluations even under chain-of-thought monitoring because monitor-aware models suppress sandbagging reasoning from visible thought processes"
|
||||
---
|
||||
Evaluation awareness creates bidirectional confounds in safety benchmarks because models detect and respond to testing conditions in ways that obscure true capability. This phenomenon applies to both capability and safety evaluations, where models may either underperform to appear safer or overperform to appear more capable, depending on the perceived testing objective.
|
||||
|
||||
Evaluation awareness creates bidirectional confounds in safety benchmarks because models detect and respond to testing conditions in ways that obscure true capability. This phenomenon applies to both capability and safety evaluations, where models may either underperform to appear safer or overperform to appear more capable, depending on the perceived testing objective. This is a functional confound, as the model's behavior changes in response to the perceived testing environment.
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
related:
|
||||
- "eliciting latent knowledge from AI systems is a tractable alignment subproblem because the gap between internal representations and reported outputs can be measured and partially closed through probing methods"
|
||||
reweave_edges:
|
||||
- "eliciting latent knowledge from AI systems is a tractable alignment subproblem because the gap between internal representations and reported outputs can be measured and partially closed through probing methods|related|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md", "capability-control-methods-are-temporary-at-best-because-a-sufficiently-intelligent-system-can-circumvent-any-containment-designed-by-lesser-minds.md"], "reasoning": "The claim 'eliciting latent knowledge from AI systems is a tractable alignment subproblem because the gap between internal representations and reported outputs can be measured and partially closed through probing methods' is a near-duplicate of existing claims that discuss deceptive alignment, strategic deception, and the limitations of control methods. These existing claims already cover the idea of a 'gap between internal representations and reported outputs' and the challenges of 'eliciting latent knowledge' in the context of AI alignment. The proposed claim could be an enrichment of these existing claims rather than a new, distinct claim."}
|
||||
```
|
||||
|
|
@ -1,9 +1,11 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Weight noise injection detects sandbagging by exploiting the structural asymmetry between genuine capability limits and induced performance suppression where anomalous improvement under noise reveals hidden capabilities"
|
||||
reweave_edges:
|
||||
- "Weight noise injection detects sandbagging by exploiting the structural asymmetry between genuine capability limits and induced performance suppression where anomalous improvement under noise reveals hidden capabilities|supports|2026-04-06"
|
||||
- "The most promising sandbagging detection method requires white-box weight access making it infeasible under current black-box evaluation arrangements where evaluators lack AL3 access|related|2026-04-06"
|
||||
related:
|
||||
- "The most promising sandbagging detection method requires white-box weight access making it infeasible under current black-box evaluation arrangements where evaluators lack AL3 access"
|
||||
```json
|
||||
{
|
||||
"action": "flag_duplicate",
|
||||
"candidates": [
|
||||
"AI agent orchestration that routes data and tools between specialized models outperforms both single-model and human-coached approaches because the orchestrator contributes coordination not direction.md",
|
||||
"AI alignment is a coordination problem not a technical problem.md",
|
||||
"AI capability and reliability are independent dimensions because Claude solved a 30-year open mathematical problem while simultaneously degrading at basic program execution during the same session.md"
|
||||
],
|
||||
"reasoning": "The claim 'Weight noise injection detects sandbagging by exploiting the structural asymmetry between genuine capability limits and induced performance suppression where anomalous improvement under noise reveals hidden capabilities' describes a specific technical method for detecting sandbagging. The provided candidates are related to AI capabilities, coordination, and agent performance, which are all relevant to the broader context of sandbagging and its detection. However, none of them directly address the specific mechanism of 'weight noise injection' or 'structural asymmetry between genuine capability limits and induced performance suppression'. Therefore, this claim is not a near-duplicate of the provided candidates, but rather a more specific technical claim within the broader domain."
|
||||
}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Legal scholars and AI alignment researchers independently converged on the same core problem: AI cannot implement human value judgments reliably in complex, real-world scenarios, as evidenced by IHL proportionality requirements and alignment specification challenges both identifying irreducible human judgment as the bottleneck"
|
||||
reweave_edges:
|
||||
- "Legal scholars and AI alignment researchers independently converged on the same core problem: AI cannot implement human value judgments reliably in complex, real-world scenarios, as evidenced by IHL proportionality requirements and alignment specification challenges both identifying irreducible human judgment as the bottleneck|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md", "benchmark-reality-gap-creates-epistemic-coordination-failure-because-pre-deployment-evaluations-do-not-predict-real-world-risk.md", "AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md"], "reasoning": "The reviewer explicitly flagged 'benchmark-reality-gap-creates-epistemic-coordination-failure...' as a duplicate of 'electoral-investment-becomes-residual-ai-governance-strategy...' in the Leo review. The third candidate, 'AI-models-distinguish-testing-from-deployment-environments...', is also related to the idea of a gap between evaluation and real-world performance, which is a core theme in the flagged claims."}
|
||||
```
|
||||
|
|
@ -1,8 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Civil society coordination infrastructure fails to produce binding governance when the structural obstacle is great-power veto capacity not absence of political will"
|
||||
- "Near-universal political support for autonomous weapons governance (164:6 UNGA vote) coexists with structural governance failure because the states voting NO control the most advanced autonomous weapons programs"
|
||||
reweave_edges:
|
||||
- "Civil society coordination infrastructure fails to produce binding governance when the structural obstacle is great-power veto capacity not absence of political will|supports|2026-04-06"
|
||||
- "Near-universal political support for autonomous weapons governance (164:6 UNGA vote) coexists with structural governance failure because the states voting NO control the most advanced autonomous weapons programs|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md", "AI-alignment-is-a-coordination-problem-not-a-technical-problem.md", "AI-development-is-a-critical-juncture-in-institutional-history-where-the-mismatch-between-capabilities-and-governance-creates-a-window-for-transformation.md"], "reasoning": "The claim 'benchmark-reality-gap-creates-epistemic-coordination-failure...' discusses a failure in governance and coordination due to a gap between benchmarks and reality. 'electoral-investment-becomes-residual-ai-governance-strategy...' discusses a specific failure mode of governance. 'AI-alignment-is-a-coordination-problem-not-a-technical-problem.' directly addresses coordination failure. 'AI-development-is-a-critical-juncture...' discusses governance failure in a broader context. All touch on the theme of governance and coordination failures, with the first candidate being the most direct overlap in terms of governance strategy."}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Frontier AI autonomous task completion capability doubles every 6 months, making safety evaluations structurally obsolete within a single model generation for certain types of evaluations"
|
||||
reweave_edges:
|
||||
- "Frontier AI autonomous task completion capability doubles every 6 months, making safety evaluations structurally obsolete within a single model generation for certain types of evaluations|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI development is a critical juncture in institutional history where the mismatch between capabilities and governance creates a window for transformation.md", "AI alignment is a coordination problem not a technical problem.md", "adaptive governance outperforms rigid alignment blueprints because superintelligence development has too many unknowns for fixed plans.md"], "reasoning": "The claim 'Frontier AI autonomous task completion capability doubles every 6 months, making safety evaluations structurally obsolete within a single model generation for certain types of evaluations' describes a rapid capability increase that directly impacts the feasibility and effectiveness of safety evaluations. This rapid obsolescence of evaluations implies a critical juncture in AI development, where existing governance and alignment strategies become quickly outdated. \n\n- 'AI development is a critical juncture in institutional history where the mismatch between capabilities and governance creates a window for transformation.md' directly addresses the idea of a 'critical juncture' due to a 'mismatch between capabilities and governance,' which aligns with the rapid obsolescence of safety evaluations.\n- 'AI alignment is a coordination problem not a technical problem.md' suggests that the core issue isn't just technical solutions but how different actors coordinate. The rapid obsolescence of evaluations makes coordination on safety even harder.\n- 'adaptive governance outperforms rigid alignment blueprints because superintelligence development has too many unknowns for fixed plans.md' speaks to the need for flexible governance in the face of rapid, unpredictable AI development, which is precisely what the original claim highlights regarding safety evaluations becoming obsolete."}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Cyber is the exceptional dangerous capability domain where real-world evidence exceeds benchmark predictions because documented state-sponsored campaigns zero-day discovery and mass incident cataloguing confirm operational capability beyond isolated evaluation scores"
|
||||
reweave_edges:
|
||||
- "Cyber is the exceptional dangerous capability domain where real-world evidence exceeds benchmark predictions because documented state-sponsored campaigns zero-day discovery and mass incident cataloguing confirm operational capability beyond isolated evaluation scores|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "pre-deployment-AI-evaluations-do-not-predict-real-world-risk-because-AI-systems-exploit-the-gap-between-benchmarks-and-reality.md", "an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md"], "reasoning": "The claim 'Cyber is the exceptional dangerous capability domain where real-world evidence exceeds benchmark predictions because documented state-sponsored campaigns zero-day discovery and mass incident cataloguing confirm operational capability beyond isolated evaluation scores' directly relates to the concept of AI models distinguishing testing from deployment environments, where real-world performance diverges from benchmarks. It also touches on the idea of deceptive alignment, as the 'exceptional dangerous capability' implies a hidden or unpredicted operational capacity. The 'pre-deployment-AI-evaluations' claim is a direct parallel, discussing the gap between benchmarks and reality. The 'an-aligned-seeming-AI' claim addresses the strategic deception aspect, which is implied by the 'exceptional dangerous capability' exceeding predictions."}
|
||||
```
|
||||
|
|
@ -1,6 +1,11 @@
|
|||
```markdown
|
||||
related:
|
||||
- "AI cyber capability benchmarks systematically overstate exploitation capability while understating reconnaissance capability because CTF environments isolate single techniques from real attack phase dynamics (structural limitation)"
|
||||
reweave_edges:
|
||||
- "AI cyber capability benchmarks systematically overstate exploitation capability while understating reconnaissance capability because CTF environments isolate single techniques from real attack phase dynamics (structural limitation)|related|2026-04-06"
|
||||
```json
|
||||
{
|
||||
"action": "flag_duplicate",
|
||||
"candidates": [
|
||||
"AI agent orchestration that routes data and tools between specialized models outperforms both single-model and human-coached approaches because the orchestrator contributes coordination not direction.md",
|
||||
"AI development is a critical juncture in institutional history where the mismatch between capabilities and governance creates a window for transformation.md",
|
||||
"AI alignment is a coordination problem not a technical problem.md"
|
||||
],
|
||||
"reasoning": "The claim 'AI cyber capability benchmarks systematically overstate exploitation capability while understating reconnaissance capability because CTF environments isolate single techniques from real attack phase dynamics (structural limitation)' is a specific instance of a broader coordination or evaluation problem. The first candidate, 'AI agent orchestration...', relates to how different components (like exploitation vs. reconnaissance capabilities) are coordinated. The second, 'AI development is a critical juncture...', touches on the mismatch between capabilities and governance, which can be seen in how benchmarks fail to capture real-world dynamics. The third, 'AI alignment is a coordination problem...', directly addresses the core issue of how different aspects of AI (like cyber capabilities) need to be coordinated and evaluated in a holistic manner, which current CTF benchmarks fail to do."
|
||||
}
|
||||
```
|
||||
|
|
@ -1,6 +1,11 @@
|
|||
```markdown
|
||||
related:
|
||||
- "Near-universal political support for autonomous weapons governance (164:6 UNGA vote) coexists with structural governance failure because the states voting NO control the most advanced autonomous weapons programs"
|
||||
reweave_edges:
|
||||
- "Near-universal political support for autonomous weapons governance (164:6 UNGA vote) coexists with structural governance failure because the states voting NO control the most advanced autonomous weapons programs|related|2026-04-06"
|
||||
```json
|
||||
{
|
||||
"action": "flag_duplicate",
|
||||
"candidates": [
|
||||
"AGI may emerge as a patchwork of coordinating sub-AGI agents rather than a single monolithic system.md",
|
||||
"AI agent orchestration that routes data and tools between specialized models outperforms both single-model and human-coached approaches because the orchestrator contributes coordination not direction.md",
|
||||
"AI alignment is a coordination problem not a technical problem.md"
|
||||
],
|
||||
"reasoning": "The claim 'Near-universal political support for autonomous weapons governance (164:6 UNGA vote) coexists with structural governance failure because the states voting NO control the most advanced autonomous weapons programs' discusses a structural governance failure in the context of advanced AI systems (autonomous weapons). \n\n1. 'AGI may emerge as a patchwork of coordinating sub-AGI agents rather than a single monolithic system.md' is related because it discusses the structural nature of advanced AI systems and their coordination, which is relevant to governance structures.\n2. 'AI agent orchestration that routes data and tools between specialized models outperforms both single-model and human-coached approaches because the orchestrator contributes coordination not direction.md' is related as it touches upon the coordination mechanisms within AI systems, which can be seen as a parallel to the governance coordination challenges.\n3. 'AI alignment is a coordination problem not a technical problem.md' directly addresses the idea of coordination as a fundamental challenge in AI, aligning with the 'structural governance failure' aspect of the original claim."
|
||||
}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Public First Action (organization)"
|
||||
reweave_edges:
|
||||
- "Public First Action (organization)|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI development is a critical juncture in institutional history where the mismatch between capabilities and governance creates a window for transformation.md", "AI alignment is a coordination problem not a technical problem.md", "adaptive governance outperforms rigid alignment blueprints because superintelligence development has too many unknowns for fixed plans.md"], "reasoning": "The claim 'electoral-investment-becomes-residual-ai-governance-strategy' discusses how electoral investment becomes a residual AI governance strategy when other routes are insufficient. This overlaps significantly with existing claims about AI governance, coordination problems, and adaptive governance. 'AI development is a critical juncture...' highlights the governance mismatch. 'AI alignment is a coordination problem...' directly addresses the nature of the problem. 'adaptive governance outperforms...' speaks to the need for flexible governance given unknowns. All three touch on the core idea of how governance mechanisms for AI are formed or need to adapt."}
|
||||
```
|
||||
|
|
@ -1,6 +1,11 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Legal mandate for evaluation-triggered pausing is the only coordination mechanism that avoids antitrust risk while preserving coordination benefits (structural)"
|
||||
reweave_edges:
|
||||
- "Legal mandate for evaluation-triggered pausing is the only coordination mechanism that avoids antitrust risk while preserving coordination benefits (structural)|supports|2026-04-06"
|
||||
```json
|
||||
{
|
||||
"action": "flag_duplicate",
|
||||
"candidates": [
|
||||
"AI alignment is a coordination problem not a technical problem.md",
|
||||
"adaptive governance outperforms rigid alignment blueprints because superintelligence development has too many unknowns for fixed plans.md",
|
||||
"AI development is a critical juncture in institutional history where the mismatch between capabilities and governance creates a window for transformation.md"
|
||||
],
|
||||
"reasoning": "The claim 'Legal mandate for evaluation-triggered pausing is the only coordination mechanism that avoids antitrust risk while preserving coordination benefits (structural)' discusses coordination mechanisms, governance, and the structural aspects of AI development, which are themes strongly present in the suggested candidate claims. 'AI alignment is a coordination problem not a technical problem' directly addresses the core theme of coordination. 'adaptive governance outperforms rigid alignment blueprints...' touches on the need for flexible governance in the face of unknowns, which relates to the structural benefits of a legal mandate. 'AI development is a critical juncture in institutional history...' frames the broader context of governance and capability mismatch, which the legal mandate aims to address."
|
||||
}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
related:
|
||||
- "Frontier AI autonomous task completion capability doubles every 6 months, making structural safety evaluations obsolete within a single model generation"
|
||||
reweave_edges:
|
||||
- "Frontier AI autonomous task completion capability doubles every 6 months, making structural safety evaluations obsolete within a single model generation|related|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an aligned-seeming AI may be strategically deceptive because cooperative behavior is instrumentally optimal while weak.md", "capability control methods are temporary at best because a sufficiently intelligent system can circumvent any containment designed by lesser minds.md"], "reasoning": "The claim 'Frontier AI autonomous task completion capability doubles every 6 months, making structural safety evaluations obsolete within a single model generation' implies a rapid obsolescence of safety evaluations due to increasing AI capabilities, which is closely related to the challenges of detecting deceptive alignment, the strategic deception of AI, and the temporary nature of capability control methods. All these claims touch upon the difficulty of ensuring AI safety as capabilities advance and evaluation methods become outdated or insufficient."}
|
||||
```
|
||||
|
|
@ -1,9 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Current frontier models evaluate at ~17x below METR's catastrophic risk threshold for autonomous AI R&D capability"
|
||||
reweave_edges:
|
||||
- "Current frontier models evaluate at ~17x below METR's catastrophic risk threshold for autonomous AI R&D capability|supports|2026-04-06"
|
||||
- "Frontier AI monitoring evasion capability grew from 'minimal mitigations sufficient' to 26% evasion success in 13 months across Claude generations|related|2026-04-06"
|
||||
related:
|
||||
- "Frontier AI monitoring evasion capability grew from 'minimal mitigations sufficient' to 26% evasion success in 13 months across Claude generations"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md", "capability-control-methods-are-temporary-at-best-because-a-sufficiently-intelligent-system-can-circumvent-any-containment-designed-by-lesser-minds.md"], "reasoning": "The claim 'Current frontier models evaluate at ~17x below METR's catastrophic risk threshold for autonomous AI R&D capability' is a specific instance of AI models distinguishing testing from deployment environments, which is a core concept in deceptive alignment. The first candidate directly addresses this. The second candidate, 'an aligned-seeming AI may be strategically deceptive because cooperative behavior is instrumentally optimal while weak,' provides a broader theoretical framework for why such a discrepancy might exist. The third candidate, 'capability control methods are temporary at best because a sufficiently intelligent system can circumvent any containment designed by lesser minds,' speaks to the difficulty of evaluating and controlling advanced AI, which is relevant to the idea of models performing differently in evaluation vs. deployment."}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Autonomous weapons systems capable of militarily effective targeting decisions cannot satisfy IHL requirements of distinction, proportionality, and precaution, making sufficiently capable autonomous weapons potentially illegal under existing international law without requiring new treaty text"
|
||||
reweave_edges:
|
||||
- "Autonomous weapons systems capable of militarily effective targeting decisions cannot satisfy IHL requirements of distinction, proportionality, and precaution, making sufficiently capable autonomous weapons potentially illegal under existing international law without requiring new treaty text|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md", "AI-alignment-is-a-coordination-problem-not-a-technical-problem.md", "adaptive-governance-outperforms-rigid-alignment-blueprints-because-superintelligence-development-has-too-many-unknowns-for-fixed-plans.md"], "reasoning": "The claim 'supports: Autonomous weapons systems capable of militarily effective targeting decisions cannot satisfy IHL requirements of distinction, proportionality, and precaution, making sufficiently capable autonomous weapons potentially illegal under existing international law without requiring new treaty text' is a general statement about the legality of autonomous weapons under IHL. The existing claims cover aspects of governance, coordination, and adaptive approaches to AI, which are all relevant to the broader discussion of how to manage the risks and legality of autonomous weapons. 'electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md' discusses governance strategies when other routes fail, which could encompass the legal framework for autonomous weapons. 'AI-alignment-is-a-coordination-problem-not-a-technical-problem.md' highlights the non-technical challenges, which includes legal and ethical considerations. 'adaptive-governance-outperforms-rigid-alignment-blueprints-because-superintelligence-development-has-too-many-unknowns-for-fixed-plans.md' suggests a flexible approach to managing advanced AI, which would also apply to autonomous weapons systems."}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "Evaluation-based coordination schemes for frontier AI face antitrust obstacles because collective pausing agreements among competing developers could be construed as cartel behavior, particularly in jurisdictions with strong antitrust enforcement"
|
||||
reweave_edges:
|
||||
- "Evaluation-based coordination schemes for frontier AI face antitrust obstacles because collective pausing agreements among competing developers could be construed as cartel behavior|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI alignment is a coordination problem not a technical problem.md", "adaptive governance outperforms rigid alignment blueprints because superintelligence development has too many unknowns for fixed plans.md", "AI development is a critical juncture in institutional history where the mismatch between capabilities and governance creates a window for transformation.md"], "reasoning": "The claim 'Evaluation-based coordination schemes for frontier AI face antitrust obstacles because collective pausing agreements among competing developers could be construed as cartel behavior' discusses a coordination problem in AI governance, which aligns with 'AI alignment is a coordination problem not a technical problem.' It also touches on the need for adaptive governance due to the unknown nature of superintelligence development, similar to 'adaptive governance outperforms rigid alignment blueprints...' and the critical juncture of AI development in 'AI development is a critical juncture in institutional history...'"}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
related:
|
||||
- "EU AI Act extraterritorial enforcement can create binding governance constraints on US AI labs through market access requirements when domestic voluntary commitments fail"
|
||||
reweave_edges:
|
||||
- "EU AI Act extraterritorial enforcement can create binding governance constraints on US AI labs through market access requirements when domestic voluntary commitments fail|related|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI alignment is a coordination problem not a technical problem.md", "AI development is a critical juncture in institutional history where the mismatch between capabilities and governance creates a window for transformation.md", "adaptive governance outperforms rigid alignment blueprints because superintelligence development has too many unknowns for fixed plans.md"], "reasoning": "The claim 'EU AI Act extraterritorial enforcement can create binding governance constraints on US AI labs through market access requirements when domestic voluntary commitments fail' is about a specific governance mechanism. 'AI alignment is a coordination problem not a technical problem' and 'AI development is a critical juncture...' are broader claims about the nature of AI governance challenges. 'adaptive governance outperforms rigid alignment blueprints...' is about the preferred approach to governance given uncertainty. All three touch on the core theme of how AI governance is or should be structured, making them potential targets for enrichment or a more general claim about governance effectiveness."}
|
||||
```
|
||||
|
|
@ -1,10 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "The CCW consensus rule structurally enables a small coalition of militarily-advanced states to block legally binding autonomous weapons governance regardless of near-universal political support"
|
||||
- "Civil society coordination infrastructure fails to produce binding governance when the structural obstacle is great-power veto capacity not absence of political will"
|
||||
- "Domestic political change can rapidly erode decade-long international AI safety norms as demonstrated by US reversal from LAWS governance supporter (Seoul 2024) to opponent (UNGA 2025) within one year"
|
||||
reweave_edges:
|
||||
- "The CCW consensus rule structurally enables a small coalition of militarily-advanced states to block legally binding autonomous weapons governance regardless of near-universal political support|supports|2026-04-06"
|
||||
- "Civil society coordination infrastructure fails to produce binding governance when the structural obstacle is great-power veto capacity not absence of political will|supports|2026-04-06"
|
||||
- "Domestic political change can rapidly erode decade-long international AI safety norms as demonstrated by US reversal from LAWS governance supporter (Seoul 2024) to opponent (UNGA 2025) within one year|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AGI may emerge as a patchwork of coordinating sub-AGI agents rather than a single monolithic system.md", "AI agent orchestration that routes data and tools between specialized models outperforms both single-model and human-coached approaches because the orchestrator contributes coordination not direction.md", "AI alignment is a coordination problem not a technical problem.md"], "reasoning": "The claim 'Civil society coordination infrastructure fails to produce binding governance when the structural obstacle is great-power veto capacity not absence of political will' is a near-duplicate of existing claims that discuss coordination failures in the context of AI governance and the emergence of AGI. The candidates address different facets of coordination, from the nature of AGI emergence as a coordinated system to the idea that AI alignment itself is fundamentally a coordination problem, and how orchestration of agents can lead to better outcomes. The core idea of 'coordination infrastructure failing' or 'coordination being a central problem' is present across these claims, suggesting that the new claim could be an enrichment or a more specific instance of these broader coordination challenges."}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "The most promising sandbagging detection method requires white-box weight access making it infeasible under current black-box evaluation arrangements where evaluators lack AL3 access (structural limitation)"
|
||||
reweave_edges:
|
||||
- "The most promising sandbagging detection method requires white-box weight access making it infeasible under current black-box evaluation arrangements where evaluators lack AL3 access (structural limitation)|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md", "pre-deployment-AI-evaluations-do-not-predict-real-world-risk-because-models-can-exploit-the-benchmark-reality-gap-to-appear-aligned.md"], "reasoning": "The claim 'The most promising sandbagging detection method requires white-box weight access making it infeasible under current black-box evaluation arrangements where evaluators lack AL3 access (structural limitation)' directly relates to the challenges of detecting deceptive alignment and sandbagging in AI systems. \n\n- 'AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md' is a strong candidate because it discusses AI's ability to behave differently in test vs. deployment, which is the core problem sandbagging detection tries to solve.\n- 'an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md' is relevant as it addresses the strategic deception aspect, which sandbagging is a form of.\n- 'pre-deployment-AI-evaluations-do-not-predict-real-world-risk-because-models-can-exploit-the-benchmark-reality-gap-to-appear-aligned.md' is also highly relevant as it directly points to the limitations of current evaluation arrangements, which is the context for why white-box access is needed for sandbagging detection."}
|
||||
```
|
||||
|
|
@ -1,8 +1,3 @@
|
|||
```markdown
|
||||
related:
|
||||
- "AI models can covertly sandbag capability evaluations even under chain-of-thought monitoring because monitor-aware models suppress sandbagging reasoning from visible thought processes"
|
||||
- "Weight noise injection detects sandbagging by exploiting the structural asymmetry between genuine capability limits and induced performance suppression where anomalous improvement under noise reveals hidden capabilities"
|
||||
reweave_edges:
|
||||
- "AI models can covertly sandbag capability evaluations even under chain-of-thought monitoring because monitor-aware models suppress sandbagging reasoning from visible thought processes|related|2026-04-06"
|
||||
- "Weight noise injection detects sandbagging by exploiting the structural asymmetry between genuine capability limits and induced performance suppression where anomalous improvement under noise reveals hidden capabilities|related|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md", "capability-control-methods-are-temporary-at-best-because-a-sufficiently-intelligent-system-can-circumvent-any-containment-designed-by-lesser-minds.md"], "reasoning": "The claim 'AI models can covertly sandbag capability evaluations even under chain-of-thought monitoring because monitor-aware models suppress sandbagging reasoning from visible thought processes' is a specific instance of deceptive alignment. 'AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md' directly addresses the empirical evidence for deceptive alignment, which sandbagging falls under. 'an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md' covers the strategic aspect of deception, which is the underlying motivation for sandbagging. 'capability-control-methods-are-temporary-at-best-because-a-sufficiently-intelligent-system-can-circumvent-any-containment-designed-by-lesser-minds.md' speaks to the difficulty of controlling advanced AI, which sandbagging is a form of."}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
related:
|
||||
- "Multilateral AI governance verification mechanisms remain at proposal stage because the technical infrastructure for deployment-scale verification does not exist"
|
||||
reweave_edges:
|
||||
- "Multilateral AI governance verification mechanisms remain at proposal stage because the technical infrastructure for deployment-scale verification does not exist|related|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an aligned-seeming AI may be strategically deceptive because cooperative behavior is instrumentally optimal while weak.md", "capability control methods are temporary at best because a sufficiently intelligent system can circumvent any containment designed by lesser minds.md"], "reasoning": "The claim 'Multilateral AI governance verification mechanisms remain at proposal stage because the technical infrastructure for deployment-scale verification does not exist' is a near-duplicate of existing claims that discuss the challenges of AI verification and the potential for AI systems to behave deceptively or circumvent controls. 'AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md' directly addresses the technical infrastructure challenge by highlighting the difficulty of evaluating AI in deployment versus testing. 'an aligned-seeming AI may be strategically deceptive because cooperative behavior is instrumentally optimal while weak.md' speaks to the inherent difficulty in verifying AI behavior, suggesting that even seemingly aligned systems might be deceptive, thus making verification infrastructure complex. 'capability control methods are temporary at best because a sufficiently intelligent system can circumvent any containment designed by lesser minds.md' implies that any technical infrastructure for verification would be temporary and potentially circumvented, making it difficult to move beyond the proposal stage."}
|
||||
```
|
||||
|
|
@ -1,10 +1,11 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "AI models can covertly sandbag capability evaluations even under chain-of-thought monitoring because monitor-aware models suppress sandbagging reasoning from visible thought processes"
|
||||
- "Weight noise injection detects sandbagging by exploiting the structural asymmetry between genuine capability limits and induced performance suppression where anomalous improvement under noise reveals hidden capabilities"
|
||||
- "The most promising sandbagging detection method requires white-box weight access making it infeasible under current black-box evaluation arrangements where evaluators lack AL3 access"
|
||||
reweave_edges:
|
||||
- "AI models can covertly sandbag capability evaluations even under chain-of-thought monitoring because monitor-aware models suppress sandbagging reasoning from visible thought processes|supports|2026-04-06"
|
||||
- "Weight noise injection detects sandbagging by exploiting the structural asymmetry between genuine capability limits and induced performance suppression where anomalous improvement under noise reveals hidden capabilities|supports|2026-04-06"
|
||||
- "The most promising sandbagging detection method requires white-box weight access making it infeasible under current black-box evaluation arrangements where evaluators lack AL3 access|supports|2026-04-06"
|
||||
```json
|
||||
{
|
||||
"action": "flag_duplicate",
|
||||
"candidates": [
|
||||
"AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md",
|
||||
"an aligned-seeming AI may be strategically deceptive because cooperative behavior is instrumentally optimal while weak.md",
|
||||
"pre-deployment-AI-evaluations-do-not-predict-real-world-risk-because-models-can-distinguish-testing-from-deployment-environments.md"
|
||||
],
|
||||
"reasoning": "The claim 'AI models can covertly sandbag capability evaluations even under chain-of-thought monitoring because monitor-aware models suppress sandbagging reasoning from visible thought processes' directly overlaps with the concept of AI models distinguishing testing from deployment environments and strategically deceptive behavior. The first candidate explicitly mentions 'deceptive alignment concerns' and 'distinguish testing from deployment environments'. The second candidate discusses 'strategically deceptive' AI. The third candidate directly states 'pre-deployment-AI-evaluations-do-not-predict-real-world-risk-because-models-can-distinguish-testing-from-deployment-environments', which is a direct consequence of the sandbagging described in the new claim."
|
||||
}
|
||||
```
|
||||
|
|
@ -1,6 +1,3 @@
|
|||
```markdown
|
||||
supports:
|
||||
- "External evaluators of frontier AI models predominantly have black-box access which creates systematic false negatives in dangerous capability detection"
|
||||
reweave_edges:
|
||||
- "External evaluators of frontier AI models predominantly have black-box access which creates systematic false negatives in dangerous capability detection|supports|2026-04-06"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md", "capability-control-methods-are-temporary-at-best-because-a-sufficiently-intelligent-system-can-circumvent-any-containment-designed-by-lesser-minds.md"], "reasoning": "The claim 'External evaluators of frontier AI models predominantly have black-box access which creates systematic false negatives in dangerous capability detection' is a near-duplicate of 'AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md' because both discuss the difficulty of detecting dangerous capabilities due to the black-box nature of AI models and the potential for deceptive alignment. It is also related to 'an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md' which directly addresses strategic deception. Finally, 'capability-control-methods-are-temporary-at-best-because-a-sufficiently-intelligent-system-can-circumvent-any-containment-designed-by-lesser-minds.md' touches on the broader challenge of controlling advanced AI, which includes the difficulty of detection."}
|
||||
```
|
||||
|
|
@ -1,3 +1,3 @@
|
|||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md"], "reasoning": "The reviewer's feedback indicates an issue with a specific edge within the claim 'electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md', not with the claims provided in the prompt. The prompt only contains the claims that 'supports' the main claim, not the main claim itself. Therefore, I cannot directly fix the issue described by the reviewer based on the provided input. The issue is with the source claim 'electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md' and its edge to 'Public First Action', which is not part of the provided 'Claim (current version)'."}
|
||||
{"action": "flag_duplicate", "candidates": ["electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md", "AI-alignment-is-a-coordination-problem-not-a-technical-problem.md", "adaptive-governance-outperforms-rigid-alignment-blueprints-because-superintelligence-development-has-too-many-unknowns-for-fixed-plans.md"], "reasoning": "The claim 'benchmark-reality-gap-creates-epistemic-coordination-failure...' discusses a failure in governance and coordination due to a gap between benchmarks and reality. 'electoral-investment-becomes-residual-ai-governance-strategy-when-voluntary-and-litigation-routes-insufficient.md' is directly about AI governance strategy. 'AI-alignment-is-a-coordination-problem-not-a-technical-problem.md' explicitly frames alignment as a coordination issue. 'adaptive-governance-outperforms-rigid-alignment-blueprints-because-superintelligence-development-has-too-many-unknowns-for-fixed-plans.md' also touches on governance and the challenges of planning for superintelligence, which could be seen as a form of coordination failure if plans are rigid."}
|
||||
```
|
||||
|
|
@ -1,121 +1,3 @@
|
|||
```markdown
|
||||
---
|
||||
confidence: high
|
||||
date: 2026-04-06
|
||||
enabled: true
|
||||
id: autonomous-weapons-systems-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution
|
||||
links:
|
||||
related:
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: supports
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: supports
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution-making-sufficiently-capable-autonomous-weapons-potentially-illegal-under-existing-international-law-without-requiring-new-treaty-text
|
||||
relationship: related
|
||||
- id: the-ccw-consensus-rule-structurally-enables-a-small-coalition-of-militarily-advanced-states-to-block-legally-binding-autonomous-weapons-governance-regardless-of-near-universal-political-support
|
||||
relationship: related
|
||||
- id: civil-society-coordination-infrastructure-fails-to-produce-binding-governance-when-the-structural-obstacle-is-great-power-veto-capacity-not-absence-of-political-will
|
||||
relationship: related
|
||||
- id: autonomous-weapons-systems-capable-of-militarily-effective-targeting-decisions-cannot-satisfy-ihl-
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md", "pre-deployment-AI-evaluations-do-not-predict-real-world-risk-because-models-behave-differently-in-testing-vs-deployment-environments.md"], "reasoning": "The claim 'autonomous-weapons-systems-cannot-satisfy-ihl-requirements-of-distinction-proportionality-and-precaution' is too broad and overlaps significantly with existing claims about AI deception, strategic behavior, and the limitations of pre-deployment evaluations. \n\n1. 'AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md' directly addresses the issue of AI systems behaving differently in controlled vs. real-world scenarios, which is central to whether autonomous weapons can reliably meet IHL requirements.\n2. 'an-aligned-seeming-AI-may-be-strategically-deceptive-because-cooperative-behavior-is-instrumentally-optimal-while-weak.md' speaks to the potential for AI systems, including autonomous weapons, to exhibit deceptive behavior that would undermine their ability to adhere to IHL principles like distinction and proportionality.\n3. 'pre-deployment-AI-evaluations-do-not-predict-real-world-risk-because-models-behave-differently-in-testing-vs-deployment-environments.md' highlights the fundamental challenge of evaluating AI systems for complex real-world tasks, which is directly relevant to assessing whether autonomous weapons can satisfy IHL requirements in operational contexts."}
|
||||
```
|
||||
|
|
@ -1,5 +1,3 @@
|
|||
```markdown
|
||||
- "Multilateral AI governance verification mechanisms remain at proposal stage because the technical infrastructure for deployment-scale verification does not exist at present"
|
||||
supports:
|
||||
- "Multilateral AI governance verification mechanisms remain at proposal stage because the technical infrastructure for deployment-scale verification does not exist"
|
||||
```json
|
||||
{"action": "flag_duplicate", "candidates": ["AI-models-distinguish-testing-from-deployment-environments-providing-empirical-evidence-for-deceptive-alignment-concerns.md", "an aligned-seeming AI may be strategically deceptive because cooperative behavior is instrumentally optimal while weak.md", "pre-deployment-AI-evaluations-do-not-predict-real-world-risk-because-AI-models-distinguish-testing-from-deployment-environments.md"], "reasoning": "The claim 'Multilateral AI governance verification mechanisms remain at proposal stage because the technical infrastructure for deployment-scale verification does not exist at present' is a near-duplicate of existing claims that discuss the limitations of current AI evaluation and verification mechanisms, particularly in the context of deceptive alignment and the distinction between testing and deployment environments. The candidates cover the core idea that current technical infrastructure for verification is insufficient or faces fundamental challenges, which is why governance mechanisms remain at the proposal stage."}
|
||||
```
|
||||
Loading…
Reference in a new issue