The AI horizon, in months.

The original reference dates are AGI in 2031, ASI in 2035 and conditional extinction in 2042. Fast takeoff starts at 2028, 2030 and 2033; slower progress at 2040, 2050 and 2065. These are editorial assumptions, not expert consensus, probabilities or statistical confidence intervals.

The independent model preserves the dates already reached when it is enabled, then applies separate evidence adjustments to each horizon. A change to AGI alone leaves ASI and conditional extinction unchanged. The illustrated pathway retains at least one month between milestones; a conflicting proposal is limited without moving its neighbors.

Twice-daily collection gathers allowlisted RSS feeds. An optional model reviews up to 24 excerpts per run with source quotations and limitations. Review is not independent fact-checking. Applying proposals requires automatic application to be enabled, recent dated evidence and a new event.

Each article gets a separate timing assessment for AGI, ASI and conditional extinction. High-confidence proposals carry full weight; medium-confidence proposals carry one-quarter weight; low-confidence proposals have no numerical effect. An article can affect one, several or none of the horizons. A cyber incident alone does not establish extinction timing.

Opposing proposals are combined. Each horizon is limited to two months per run and five years from its preserved starting date. Fractional-month changes can accumulate. Publisher and event limits reduce repeat counting. These are engineering choices, not statistically calibrated evidence weights.

Months remaining are approximate calendar months from today, rounded to the nearest whole month on the forecast cards. Calculations keep fractional months, and history shows changes to one decimal place. The displayed precision does not imply forecast certainty. Whole-year assumptions use January of that year. They are scenario illustrations, not event deadlines. A countdown reaching zero does not establish that an event occurred.

The history chart measures how far each predicted date moved from the first saved forecast in the selected period. An unchanged forecast stays flat as time passes. Each point comes from a saved collection run; the chart does not invent earlier predictions. Switching scenarios recalculates the chart for that assumption set.

No validated method converts current news into a numerical probability of human extinction. The previous version’s percentage and fixed countdown were unsupported and have been removed. A faster timeline is not the same thing as a higher probability of extinction.

RSS coverage is incomplete; feeds may lag or fail. Automated analysis sees excerpts, not full articles, and can misinterpret them or miss events. arXiv papers are not necessarily peer reviewed. Developer benchmarks need independent corroboration. Unforeseen breakthroughs, plateaus, policy changes and non-ASI pathways are outside this simple model. No finite range excludes “never.”