Leopold Aschenbrenner - Situational Awareness Timeline
Yearly Predictions
2025/2026:
- AI will drive $100B+ annual revenues for big tech companies
- AI will outcompete PhDs in raw problem-solving smarts
- We’ll have $10T companies (OpenAI @ $157 Billion)
2027/2028:
- We’ll have models trained on the $100B+ cluster
- Full-fledged AI agents/drop-in remote workers will start to widely automate software engineering and other cognitive jobs
Test Time Compute
| Number of tokens | Equivalent to me working on something for… | OOMs | Progress |
|---|---|---|---|
| 100s | A few minutes | ChatGPT (we are here) | ✅ |
| 1,000s | Half an hour | +1 OOMs test-time compute | ✅(OpenAI's O1-preview thinks for several minutes) |
| 10,000s | Half a workday | +2 OOMs | ⏳ |
| 100,000s | A workweek | +3 OOMs | ⏳ |
| Millions | Multiple months | +4 OOMs | ⏳ |
Training Compute
Observing the increase in model sizes and parameter counts to evaluate progress in AI capability.
| Year | OOMs | H100s-equivalent | Cost | Power | Power reference class | Progress |
|---|---|---|---|---|---|---|
| 2022 | ~GPT-4 cluster | ~10k | ~$500M | ~10 MW | ~10,000 average homes | ✅ |
| ~2024 | +1 OOM | ~100k | $billions | ~100 MW | ~100,000 homes | ✅ (xAI Mephis Datacenter, Colossus in 2024) |
| ~2026 | +2 OOMs | ~1M | $10s of billions | ~1 GW | The Hoover Dam, or a large nuclear reactor | 🚧(OpenAI Abilene Datacenter, eta mid 2026) |
| ~2028 | +3 OOMs | ~10M | $100s of billions | ~10 GW | A small/medium US state | 🚧(OpenAI + Microsoft, eta 2028) |
| ~2030 | +4 OOMs | ~100M | $1T+ | ~100 GW | >20% of US electricity production | ⏳ |
Source: Situational Awareness