METR updates its time horizon plot to add judgment-heavy tasks, finding some AI capabilities still grow exponentially but models lag at designing AI evaluations (Grace Huckins/MIT Technology Review)
The updated framework tracks and benchmarks artificial intelligence performance across increasingly complex metrics. Researchers expanded their analysis by incorporating tasks that require substantial human-like judgment into the time horizon plot. While the data confirms that certain capabilities maintain an exponential rate of expansion, the models continue to lag specifically in designing AI evaluations. This dual-track progression outlines the precise frontiers where automated capabilities remain notably constrained.
Key points
Complex tracking efforts show that artificial intelligence performance is progressing une…: Complex tracking efforts show that artificial intelligence performance is progressing unevenly across metrics.
Systems that excel at exponential tasks still struggle to perform human-like judgment eva…: Systems that excel at exponential tasks still struggle to perform human-like judgment evaluations.
Benchmark analyses help identify critical boundaries where automated machine capabilities…: Benchmark analyses help identify critical boundaries where automated machine capabilities remain constrained.
Sources checked by our analysis pipeline: 1
New York Times: |
Ashley Capoot / CNBC: |
Joseph Howlett / Scientific American: |
Katie Paul / Reuters: |
William Sandlund / Financial Times: |
Ben Schoon / 9to5Google: |
CNBC: |
Courtney Rozen / Reuters: |
Axios: |
Bloomberg: |
Katie Nicora / CBS News: |
Barry Schwartz / Search Engine Roundtable: |
Harper Harwell / Birmingham Business Journal: |
Yoolim Lee / Bloomberg: |
Olivia Carville / Bloomberg: |
Maria Deutscher / SiliconANGLE: |
Bloomberg: |
Tim Genewein / DeepMind Institute: |
Bloomberg: |
Grace Huckins / MIT Technology Review: |
Julianna Bragg / Axios: |
New York Times: |
Wall Street Journal: |
This is a Techmeme archive page. It shows how the site appeared at 2:05 PM ET, November 20, 2026.
The most current version of the site as always is available at our home page. To view an earlier snapshot click here and then modify the date indicated.
Speaking to reporters after returning from Asia, Donald Trump asserted that keeping infrastructure domestic is necessary to prevent China from seizing facilities and eclipsing American technological dominance. The proposed restrictions and threatened tax penalties would directly disrupt major international expansion plans pursued by American technology giants. However, the proposal faces complex domestic challenges, as local resistance to the facilities' massive energy consumption and noise footprint continues to…
Mathematician Levent Alpöge, an employee at Anthropic, has announced a Lean-compilable solution to the Riemann Hypothesis, a renowned problem dating back to 1859 concerning the structure underlying prime numbers. Alpöge utilized the AI model Claude to assemble the pieces of the argument, resulting in a computer-verified proof in the open-source Lean language. The announcement coincided with Anthropic becoming the first publicly traded frontier AI lab, as well as a separate proof released by Alpöge and NYU mathemat…
According to a briefing on his Friday press conference, President Donald Trump pledged punitive measures aimed at keeping critical technology infrastructure on American soil. Trump warned that moving artificial intelligence facilities abroad risks losing technological dominance to China, pointing to past losses suffered in the domestic semiconductor chip manufacturing sector. The move seeks to counteract emerging domestic regulatory barriers that have made local development increasingly difficult for tech firms.