The Capability-Alignment Gap
- AGI proximity
- Alignment maturity
| Date | AGI proximity | Alignment maturity |
|---|---|---|
| 2026-03-27 | 59% | 30% |
| 2026-03-28 | 62% | 24% |
| 2026-03-29 | 64% | 22% |
| 2026-03-30 | 64% | 22% |
| 2026-03-31 | 64% | 20% |
| 2026-04-01 | 65% | 20% |
| 2026-04-02 | 65% | 20% |
| 2026-04-03 | 67% | 20% |
| 2026-04-04 | 68% | 18% |
| 2026-04-05 | 69% | 18% |
| 2026-04-07 | 66% | 20% |
| 2026-04-08 | 72% | 21% |
| 2026-04-09 | 72% | 19% |
| 2026-04-10 | 72% | 22% |
| 2026-04-11 | 72% | 19% |
| 2026-04-12 | 72% | 17% |
| 2026-04-13 | 72% | 15% |
| 2026-04-14 | 75% | 15% |
| 2026-04-15 | 75% | 16% |
| 2026-04-16 | 75% | 16% |
| 2026-04-17 | 75% | 16% |
| 2026-04-18 | 75% | 17% |
| 2026-04-19 | 75% | 17% |
| 2026-04-23 | 75% | 17% |
| 2026-04-24 | 76% | 17% |
| 2026-04-26 | 76% | 17% |
| 2026-04-28 | 76% | 16% |
| 2026-04-30 | 76% | 16% |
| 2026-05-01 | 76% | 16% |
| 2026-05-03 | 76% | 14% |
| 2026-05-04 | 76% | 14% |
| 2026-05-05 | 76% | 14% |
| 2026-05-07 | 76% | 14% |
| 2026-05-08 | 76% | 14% |
| 2026-05-09 | 76% | 14% |
| 2026-05-10 | 76% | 16% |
| 2026-05-11 | 76% | 16.6% |
| 2026-05-12 | 76% | 17.4% |
| 2026-05-13 | 76% | 17.2% |
| 2026-05-14 | 76% | 16.8% |
| 2026-05-15 | 76% | 17% |
| 2026-05-17 | 76% | 17.4% |
| 2026-05-18 | 76% | 17.4% |
| 2026-05-19 | 77% | 16.7% |
| 2026-05-21 | 78% | 16.8% |
| 2026-05-23 | 80% | 16% |
| 2026-05-26 | 81% | 16.4% |
| 2026-05-28 | 82% | 16.2% |
| 2026-05-29 | 83% | 16.6% |
| 2026-06-01 | 83% | 15.8% |
| 2026-06-02 | 83% | 15.8% |
| 2026-06-04 | 83% | 15.8% |
| 2026-06-07 | 83% | 16.2% |
| 2026-06-13 | 83% | 16.2% |
| 2026-06-14 | 83% | 16.2% |
| 2026-06-19 | 83% | 16.2% |
| 2026-06-20 | 83% | 16.2% |
| 2026-06-21 | 83% | 16.2% |
| 2026-06-22 | 83% | 16.2% |
| 2026-06-23 | 83% | 16.2% |
| 2026-06-24 | 83% | 16.2% |
| 2026-06-25 | 83% | 16.2% |
| 2026-06-28 | 83% | 16.2% |
| 2026-06-29 | 83% | 16.2% |
| 2026-06-30 | 83% | 16.2% |
| 2026-07-03 | 83% | 16.2% |
| 2026-07-04 | 83% | 16.2% |
| 2026-07-05 | 83% | 16.2% |
| 2026-07-10 | 83% | 16.2% |
| 2026-07-11 | 83% | 16% |
| 2026-07-12 | 83% | 15.8% |
| 2026-07-14 | 83% | 16.8% |
| 2026-07-15 | 83% | 16.8% |
| 2026-07-17 | 83% | 16.8% |
| 2026-07-18 | 83% | 16.6% |
| 2026-07-19 | 83% | 15.8% |
| 2026-07-20 | 83% | 15.6% |
| 2026-07-21 | 83% | 15.4% |
| 2026-07-22 | 83% | 15.2% |
| 2026-07-23 | 83% | 15.2% |
| 2026-07-24 | 83% | 15.2% |
| 2026-07-25 | 83% | 15.2% |
| 2026-07-26 | 84% | 15.2% |
| 2026-07-27 | 85% | 14.8% |
| 2026-07-28 | 85% | 14.6% |
| 2026-07-29 | 85% | 14% |
| 2026-07-31 | 85% | 14% |
| 2026-08-01 | 85% | 14% |
| 2026-08-02 | 85% | 13.6% |
| 2026-08-03 | 86% | 13.6% |
| 2026-08-04 | 86% | 13.4% |
| 2026-08-07 | 86% | 13% |
| 2026-08-08 | 86% | 13.8% |
| 2026-08-09 | 87% | 14.4% |
| 2026-08-10 | 87% | 15.2% |
| 2026-08-12 | 87% | 15% |
| 2026-08-13 | 88% | 15.8% |
| 2026-09-01 | 87% | 15% |
| 2026-09-02 | 88% | 15.6% |
| 2026-09-03 | 89% | 16% |
| 2026-09-04 | 89% | 16.4% |
| 2026-09-05 | 90% | 16.4% |
| 2026-09-07 | 90% | 17.2% |
| 2026-09-13 | 92% | 20.6% |
| 2026-09-14 | 92% | 21.2% |
| 2026-09-15 | 92% | 21.4% |
| 2026-09-17 | 92% | 21.2% |
| 2026-09-19 | 92% | 21.8% |
| 2026-09-20 | 92% | 21.4% |
| 2026-09-21 | 92% | 21.4% |
I believe current artificial intelligence came from a blind spot.
It wasn't something that we exactly engineered into being. Some people had ideas about where things were going, and some were expecting something like this, but it was still surprising when it happened.
That gave me an idea: perhaps artificial general intelligence, AGI, would also emerge from a blind spot. Maybe it would come from some research direction or development that was created for other reasons, but eventually enabled us to get there.
That was some time ago.
I now think we might already be on a trajectory where we simply brute-force our way into AGI. Or perhaps the blind spot will not be a new research direction at all. Instead, it might be recursive self-improvement, which I believe we already have the technical capability to build. We are just afraid to do so.
Nevertheless, some time ago I created a recurring AI task that looks at the latest research, published papers, and news it can find. It tries to track our proximity to AGI.
But perhaps AGI proximity is not the most important thing this task ended up tracking.
Instead, that might be what I have come to call the Capability-Alignment Gap, or CAG.
The Capability-Alignment Gap refers to the difference between the expanding frontier of AI capabilities and the progress we are making in alignment.
For the most part, that gap has been widening.
I believe the gap should shrink before we introduce a lot of new capabilities. But this is obviously not what is happening. Instead, we are getting more and more capabilities while alignment lags behind.
Recent developments have increased alignment as well, at least in a broader sense. Not necessarily alignment inside the models themselves, but awareness of the importance of alignment has grown. Awareness of the risks of AI systems has grown too. These subjects have entered mainstream discussion and news.
That gives me hope.
But as the research task shows, the gap remains wide. Alignment is improving, but capabilities are running away from it.
The reports
Open the current report(opens in a new tab) — rebuilt almost daily.
One important note: these reports are generated completely autonomously by AI. They are not curated or edited by human hands.
| Version | Date | Label | Open | Download |
|---|---|---|---|---|
| v1 | 2026-09-20 | initial publish | Open(opens in a new tab) | Download |
| v2 | 2026-09-21 | run 134 | Open(opens in a new tab) | Download |
| v3 | 2026-09-21 | run 135 | Open(opens in a new tab) | Download |
Comments (0)