The approach
A forecast that can change its mind.
Follow the evidence. Keep the uncertainty. Publish when the assessment changes.
What the watch follows
The AI Singularity Forecast Watch tracks the path to transformative AI: systems capable of changing the pace of research, the economy, and the way society works. It considers capabilities and reliability, AI-assisted AI research, science, robotics, compute and energy, alignment, misuse, and governance.
What counts as a material change?
New evidence must change the timing, expected transition, severe-loss-of-control assessment, physical-world lag, or the confidence and reasons behind those judgments. A correction to a substantive error also qualifies. Another launch announcement or benchmark gain does not automatically qualify.
How to read an update
- Evidence
- What was observed or reported, with source links and relevant dates.
- Interpretation
- Why the evidence matters, the strongest counterevidence, and the limits of the inference.
- Revision
- What changed from the preceding assessment, and what remained unchanged.
- Uncertainty
- Company claims, estimates, experiments and real deployments are identified separately. Forecast probabilities are informal judgments, not measured frequencies.
Checking and publishing
The requested checking schedule is daily at 9:00 a.m. America/New_York. A check is not a promise of a new article. Unchanged assessments do not create posts or routine notifications. Source or analysis failures do not count as “no change.”
New analysis is required to use GPT-6.1 Sol at xhigh reasoning effort. The publishing workflow checks the recorded runtime settings and stops if they do not match; it does not silently substitute another model.
The historical archive
The original watch is preserved from its initial July 2026 forecasts through October 19. The archive includes substantive updates, corrections, original citation destinations, tables, and recovered illustrated source cards. Each report keeps its own estimates; a value omitted from a report is not inferred from later assessments. Separate archive notes identify uncertain dates, recovery gaps and later source verification.
Historical imports are excluded from new feed announcements and do not advance the research cutoff. Importing a report is not a new endorsement of its forecast. The archive includes early inconsistencies and subsequent corrections as part of the original record.
Images and appearance
Historical images are the originals shown with their reports. New reports may include informative source charts, diagrams or photographs; images are optional. Source links and captions remain available if an image cannot be displayed. Observatory and Journal show the same record. The appearance choice is stored only in this browser.