Question 01
Can the frontier keep building larger AI systems?
Tracks the compute, capital, and physical capacity that makes future training and inference possible.
Partly measured
Explore this question3 evidence lanes
5 measurement streams directly or supportingly inform this question. Open one to see its reality line and point-level evidence.
Named published views
attributed18 attributed claims from 5 registered works. They remain separate from observed reality and from one another.
Aggregate expectations
group beliefNo like-for-like panel, survey, or forecasting-platform snapshot is registered for this question.
What would change the picture: A sustained break in frontier-compute growth, or verified energized power and chip capacity that contradicts announced build plans.
Question 02
Are systems becoming more capable on real, long-horizon tasks?
Favors task completion and independently evaluated autonomy over broad capability labels.
Measured
Explore this question3 evidence lanes
1 measurement stream directly or supportingly inform this question. Open one to see its reality line and point-level evidence.
Named published views
attributed2 attributed claims from 2 registered works. They remain separate from observed reality and from one another.
Aggregate expectations
group beliefRelevant panels or forecasting platforms are registered, but no number is shown until the exact wording, sample, date, and aggregation method match.
What would change the picture: Replicated, methodology-stable evaluations showing either durable long-horizon gains or a flattening after controls for evaluation gaming.
Question 03
Is AI accelerating the creation of better AI?
Direct public evidence is limited: company reports show task-specific contributions, while independent causal evidence remains adjacent to frontier AI R&D.
Partly measured
Explore this question3 evidence lanes
No direct or supporting recurring measurement series is registered yet. The gap is being shown rather than filled with a proxy. 2 related streams are registered only as proxy or context.
Named published views
attributed6 attributed claims from 6 registered works. They remain separate from observed reality and from one another.
Aggregate expectations
group beliefVerified, exactly matched aggregate snapshots are displayed separately from observed reality and named views.
What would change the picture: A repeated, independently audited measure of AI contribution to frontier research, plus verified successor-system work with limited human direction.
Question 04
Can powerful systems be controlled, monitored, and reliably evaluated?
Tracks the hard-to-measure gap between a model appearing safe in evaluation and remaining controllable in deployment.
Open question
Explore this question2 evidence lanes
No direct or supporting recurring measurement series is registered yet. The gap is being shown rather than filled with a proxy.
Aggregate expectations
group beliefRelevant panels or forecasting platforms are registered, but no number is shown until the exact wording, sample, date, and aggregation method match.
What would change the picture: Independent safety evaluations, robust behavioral monitoring, and evidence that mitigations hold under realistic deployment pressure.
Question 05
Are algorithms improving capability faster than hardware scales?
Separates capability gains from additional hardware and records the vintage of the underlying methodology.
Partly measured
Explore this question2 evidence lanes
1 measurement stream directly or supportingly inform this question. Open one to see its reality line and point-level evidence.
Named published views
attributed3 attributed claims from 3 registered works. They remain separate from observed reality and from one another.
What would change the picture: A current, transparent fixed-capability efficiency series or evidence that observed improvements no longer transfer across tasks.
Question 06
How much capable AI is actually being deployed and used?
Distinguishes spending and revenue signals from verified autonomous deployment and economic use.
Partly measured
Explore this question3 evidence lanes
1 measurement stream directly or supportingly inform this question. Open one to see its reality line and point-level evidence. 1 additional stream is shown only as proxy or context.
Named published views
attributed7 attributed claims from 4 registered works. They remain separate from observed reality and from one another.
Aggregate expectations
group beliefNo like-for-like panel, survey, or forecasting-platform snapshot is registered for this question.
What would change the picture: Audited adoption, utilization, and productivity series that can be separated from vendor revenue and announced investment.
Question 07
Does greater capability translate into harmful real-world power?
Tests whether cognitive performance converts into practical cyber, bio, persuasion, or resource-acquisition advantage.
Open question
Explore this question2 evidence lanes
No direct or supporting recurring measurement series is registered yet. The gap is being shown rather than filled with a proxy.
Aggregate expectations
group beliefRelevant panels or forecasting platforms are registered, but no number is shown until the exact wording, sample, date, and aggregation method match.
What would change the picture: Independently documented incidents or controlled evaluations that establish a reliable capability-to-impact link, including effective defenses.
Question 08
Can institutions impose and enforce meaningful constraints?
Measures enforceable action rather than policy announcements, and keeps coordination as an open question.
Partly measured
Explore this question3 evidence lanes
1 measurement stream directly or supportingly inform this question. Open one to see its reality line and point-level evidence.
Named published views
attributed3 attributed claims from 1 registered work. They remain separate from observed reality and from one another.
Aggregate expectations
group beliefRelevant panels or forecasting platforms are registered, but no number is shown until the exact wording, sample, date, and aggregation method match.
What would change the picture: Verified implementation and enforcement evidence across relevant jurisdictions, especially where commercial incentives conflict with restraint.
Question 09
Are influential trajectory forecasts surviving contact with reality?
Compares dated, attributed claims with their intended measurements while preserving scenarios, conditions, and revisions.
Partly measured
Explore this question3 evidence lanes
No direct or supporting recurring measurement series is registered yet. The gap is being shown rather than filled with a proxy.
Named published views
attributed30 attributed claims from 8 registered works. They remain separate from observed reality and from one another.
Aggregate expectations
group beliefRelevant panels or forecasting platforms are registered, but no number is shown until the exact wording, sample, date, and aggregation method match.
What would change the picture: More comparable resolved probabilistic forecasts, pre-registered scoring rules, and direct measures for currently proxy-scored claims.
Question 10
Are real-world harms and near misses increasing?
A deliberately visible coverage gap: incidents and near misses need a curated, independently verifiable series before they can be trended.
Missing a defensible series
Explore this question2 evidence lanes
No direct or supporting recurring measurement series is registered yet. The gap is being shown rather than filled with a proxy.
Aggregate expectations
group beliefNo like-for-like panel, survey, or forecasting-platform snapshot is registered for this question.
What would change the picture: A transparent incident taxonomy with primary documentation, severity criteria, and explicit coverage limits.