Experimental dashboard · 3 versions
AGI Probability Barometer
Explore a hiring-based indicator and the assumptions behind it.
Open Grokbot Edition Compare all versionsWhat is AGI Probability Barometer?
The AGI Probability Barometer is an experimental visualization of a hiring-based formula. Its name is historical: the displayed percentages are not validated probabilities of artificial general intelligence. The app uses fixed numbers stored in its source code, rather than a live job-listings feed.
How to use it
- Open either edition to inspect the lab gauges and the stored quarterly inputs.
- The Flash edition adds a combined gauge, a trend chart and interactive lab details.
- Read the methodology below before interpreting any percentage. The two editions use different time windows and formulas.
Before you begin
A fall in job listings could reflect budgets, recruiting cycles, restructuring or many other causes. It does not demonstrate AGI.
No original data-source citations or collection method accompany the stored figures. Treat them as unverified example inputs, not current hiring statistics.
Choose a version
Each link opens that edition directly. These are descriptions of the implementations in this project, not rankings of the underlying AI models.
| Version | What to expect |
|---|---|
| Grokbot Edition | Live job-listing counts with source links and dates, one hiring-drop index formula, a dashed live-count segment on the chart and a no-data state for labs without a verified count. |
| Gemini 3.5 Flash (High) | Three stored quarterly inputs per lab, a combined indicator, trend chart and interactive formula details. |
| Gemini 3.1 Pro (High) | The earlier two-quarter dashboard, with one gauge per lab and a simpler decline-based formula. |
Methodology and limitations
The Grokbot Edition uses one formula for every lab and for the combined total: max(0, 1 − latest / peak), called the hiring-drop index. It loads live listing counts from listings.json, each with a source URL and retrieval time. Labs without a verified latest count show “no data” and are left out of the combined index. Earlier Q1–Q3 points come from the older editions and remain unverified.
The Gemini 3.1 Pro edition calculates max(0, 100 × (1 − Q2 / Q1)) for each lab. The Flash edition uses max(0, 100 × (1 − Q3 / max(Q1, Q2, Q3))). Its combined gauge applies the same decline-from-peak idea to the quarterly totals.
These formulas express a percentage decline in the stored inputs. They are not calibrated against AGI outcomes, do not establish causation and do not estimate a reliable arrival date.
The Flash screen originally labelled a June 22, 2026 cutoff while including Q3 2026 values. That timing is inconsistent with a completed Q3 observation. The values remain unchanged for reproducibility; their provenance and timing are unverified.
Stored inputs, not a live feed
| Lab label | Q1 | Q2 | Q3 (Flash) |
|---|---|---|---|
| OpenAI | 452 | 660 | 712 |
| Anthropic | 309 | 431 | 374 |
| DeepMind | 102 | 92 | 25 |
| xAI | 285 | 247 | 211 |
Before using this as a research dashboard, the dataset needs dated source URLs, an explanation of how listings were counted, consistent observation windows and a way to distinguish open roles from actual hires. Neither version currently supplies those things.
Free to try in your browser
No installation or account is needed. The interactive experience needs JavaScript. Browser storage and network access may be used for preferences, records or shared features. Read about the project and its data use.