3
TowardSingularity
// The challenge
AI progress is often caught between hype and dismissal, while most forecasts reduce a complex future to a single date. We built something more transparent: one measurable indicator per domain, a verifiable source for every number, and a clear view of what we know, what we don’t, and what remains uncertain.
// What we built
What we built, piece by piece
- 01
A dated trend wherever the data supports one
- 02
A Python pipeline that fetches, cleans and fits the data
- 03
A scoreboard that grades public predictions as they fall due
- 04
New numbers and posts published without a rebuild
Built with Nuxt, Node.js, Python and Tailwind CSS.
// The build
How we built it
There is no composite score. Training compute, inference cost, the top LMArena rating and a Manifold market are each shown in their own units, and a domain's arrival year reads only its own trend and its own bar. Software is measured on METR's task horizons, scientific reasoning on GPQA Diamond and mathematics on FrontierMath Tier 4. Each date is a probability curve from 5,000 draws, flagged when the latest reading is already within test noise of the bar.
No source can stop a run. If one is down, the fetch falls back to documented values and logs it, and raw responses are kept so any number can be traced to what the source returned. We also checked each source's terms. The forecast moved from Kalshi to Manifold because Kalshi's terms rule out building indexes, and its US-only accounts meant a reader in India could not open the source to check it.
A History page lists each pipeline run since 19 August 2026, unsmoothed, so a quoted figure can be traced to its row. Domains without a defensible series say so: computer use, robotics and self-driving name the series they still need, while medicine, law and finance have no independent measurement, and the site will not invent one.
WebsiteAl Baseo Trading
Al Baseo Trading FZE LLC · Ajman, UAE
Read more: Al Baseo Trading3languages, with Arabic right to left
