SectionsFoundationsFrom GPT-4 to AGI: counting the OOMsThe intelligence explosionThe trillion-dollar clusterSecurity and superalignmentGeopolitics and The ProjectStudy modeFoundationsExpand all01What does "situational awareness" refer to in Aschenbrenner's title?02When does Aschenbrenner predict AGI?03When does Aschenbrenner predict superintelligence, and how does he describe the 2030s?04What is an OOM?05What's the rhetorical move behind "count the OOMs"?06How does Aschenbrenner define AGI vs. superintelligence?From GPT-4 to AGI: counting the OOMsExpand all01What share of the GPT-2 → GPT-4 jump came from raw compute scaling?02What share came from algorithmic efficiency?03What's the third driver of AI progress, beyond raw compute and algorithmic efficiency?04Why does Aschenbrenner think unhobbling is the most underrated driver?05What is the "drop-in remote worker" vision?06How many OOMs of compute does Aschenbrenner forecast from GPT-4 to 2027?07How many OOMs of algorithmic efficiency does Aschenbrenner forecast from GPT-4 to 2027?08What's the total effective OOM scaleup from GPT-4 to 2027?09What is the data wall, and why doesn't Aschenbrenner think it stops progress?The intelligence explosionExpand all01Define the intelligence explosion as Aschenbrenner gives it.02By 2027, how many AGI-level agents could a frontier lab run in parallel?03How fast does each agent think relative to a human?04What is the "decade in a year" compression claim?05Bottleneck 1: limited experiment compute. Why doesn't it stop the intelligence explosion?06Bottleneck 2: complementarities and the long tail of hard-to-automate tasks. Why isn't it decisive against the intelligence explosion?07Bottleneck 3: diminishing returns to research effort. Why doesn't it stop the intelligence explosion?08What's the difference between quantitative and qualitative superintelligence?The trillion-dollar clusterExpand all01What's the 2022 cluster baseline (GPT-4 era)?02What cluster scale does Aschenbrenner project for 2026 (chips, dollars, power)?03What cluster scale does Aschenbrenner project for 2028 (chips, dollars, power)?04What cluster scale does Aschenbrenner project for 2030 (chips, dollars, power)?05Why is power, not chips, the binding constraint past 2028?06What's the headline AI capex projection by 2030?07How does the projected ~$8T/year AI capex mobilization compare to historical buildouts?08Why does it matter where the trillion-dollar cluster gets built?Security and superalignmentExpand all01What are the two "crown jewels" frontier labs need to protect?02Of the two AI lab crown jewels (model weights and algorithmic secrets), which is more underrated, and why?03How does Aschenbrenner characterize current lab security?04What level of security does Aschenbrenner think frontier AI labs need?05What is superalignment?06Why does RLHF stop working past human level?07Name the main scalable-oversight bets.08What is defense in depth, and why is it needed on top of alignment?Geopolitics and The ProjectExpand all01What is the "decisive military advantage" argument?02What's the "lead-as-safety-margin" claim?03What are the two pathways by which China could win the AGI race?04What is "The Project"?05When does Aschenbrenner predict The Project crystallizes?06Why does Aschenbrenner think USG absorption of frontier AI labs is inevitable?07What does "this decade or bust" mean?