SUPERSTRATEGY – probabilistically optimised, frontier-forecasted decisions.

Question: Will a majority of the UK public support stronger AI regulation in the latest available survey before 1 Jan 28?

Resolves: TRUE if the most recent such survey before the date shows majority support for stronger AI regulation; otherwise FALSE.

Question: Will authoritative analysis estimate that AI-related labour displacement worsened UK public finances by at least 1% of GDP in 2030?

Resolves: Will resolve as true if, by 01 Jan 33, the OBR, HM Treasury, IFS or two independent transparent analyses estimate that the net AI-related labour-market effect worsened the 2030 UK fiscal balance by at least 1% of GDP relative to a stated counterfactual. It will otherwise resolve as false.

Question: Will a frontier AI lab publicly claim to have deployed a substantially autonomous 'AI researcher' conducting AI R&D before 1 Jan 29?

Resolves: TRUE if, before 1 Jan 2029, a frontier AI lab states in its own published material that it is using an AI system as a substantially autonomous researcher in its AI research and development — meaning the lab characterises the system as selecting its own approaches, designing and running its own experiments, and iterating on results without per-step human direction — and describes this as ongoing use within its research process rather than a bounded experiment. Does not count: capability details appearing incidentally in material whose subject is something else, such as a safety incident, an evaluation result, or a benchmark record; long-running autonomous execution of a human-specified task against a pre-set metric; one-off experiments, demonstrations, or evaluation runs; systems the lab describes as assisting, accelerating, or augmenting its researchers; media characterisations not matched by a statement from the lab. Otherwise FALSE.

Question: Will the cost of at least one frontier AI training run exceed US$1 billion before 1 Jan 28?

Resolves: TRUE if Epoch AI or a comparable credible source estimates a single frontier training run above $1bn before the date; otherwise FALSE.

Why it resolved: Wccftech reported that GPT-6 Astra was trained on a 100,000-GPU cluster and estimated four cost scenarios. Its upper scenario—100,000 GPUs × 24 hours × 120 days × $3.50 per GPU-hour—was explicitly calculated at $1.008 billion. This is an estimate for one completed frontier-model training run, exceeds $1 billion, and was published after the question opened and before the deadline. https://wccftech.com/openais-1-billion-gpt-6-astra-heralds-the-age-of-agi-according-to-nvidias-jensen-huang-yet-anthropics-fable-5-1-still-beats-it-on-swe-bench-pro-and-coding-benchmarks/