This is a link post.

(written for a Twitter audience)

Has AI progress slowed down? I’ll write some personal takes and predictions in this post.

The main metric I look at is METR's time horizon, which measures the length of tasks agents can perform. It has been doubling for more than 6 years now, and might have sped up recently.

By measuring the length of tasks AI agents can complete, we can get a continuous metric of AI capabilities.

Since 2019, the time horizon has been doubling every 7 months. But since 2024, it's been doubling every 4 months. What if we irresponsibly extrapolated these to 2030?

If AI progress continues at its recent rate, we get AI systems which can do one month (167 hours) of low-context SWE work by the end of 2027. If AI progress continues at the long-run historical rate, we get them by [...]

---

First published:
August 11th, 2025

Source:
https://www.lesswrong.com/posts/GAJbegsvnd85hX3eS/thoughts-on-extrapolating-time-horizons

Linkpost URL:
https://x.com/nikolaj2030/status/1954248757513720297

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Graph showing time-horizon of software engineering tasks for different LLMs, 2020-2025

Graph showing AI model task completion times from 2019-2027, with doubling rates.

Graph showing AI model task completion times from 2019-2030, by METR.

The graph plots various AI models' ability to autonomously complete software tasks, measuring success at 50% completion rate. It shows an upward trend with two projected doubling time trajectories.

Graph showing AI model task completion times from 2019-2030, with exponential trend lines.

The visualization plots various AI models like Claude and GPT series against their ability to autonomously complete software tasks, measured in time units from seconds to hours. Two doubling time projections are shown in purple and green.

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

[Linkpost] “Thoughts on extrapolating time horizons” by Nikola Jurkovic

LessWrong (30+ Karma)