Matthew Barnett

Timeline

1. By 2023

   Researching AI trends at Epoch

   He earns a degree in computer science from UC Berkeley. By January 2023, he researches trends related to AI at Epoch AI.

2. January–July 2023

   Bets with Bryan Caplan

   In January 2023, he bets economist Bryan Caplan that an AI will earn an A on at least five of Caplan's six most recent midterm exams by January 2029. GPT-4 scores 73 out of 100 on Caplan's labor economics midterm in March 2023, an A. In July 2023, the two make a second bet on whether real gross world product will exceed 130% of its previous yearly peak by January 2043.

3. April–May 2023

   The Direct Approach

   In April 2023, he and Tamay Besiroglu publish the Direct Approach at Epoch AI, a way to forecast AI timelines by extrapolating and interpreting the loss of machine learning models as described by scaling laws. His May 2023 essay "A compute-based framework for thinking about the future of AI" argues that AI's potential to automate labor is likely to alter the course of human history within decades, with compute as the main driver of progress.

4. April–August 2024

   Chinchilla and transfer scaling laws

   In April 2024, he co-authors "Chinchilla Scaling: A replication attempt" with Tamay Besiroglu, Ege Erdil and Josh You, which finds the scaling-law estimates reported by Hoffmann et al. inconsistent with their first two estimation methods. In August 2024, he publishes an empirical study of scaling laws for transfer learning in transformer models.

5. November 2024

   The FrontierMath benchmark

   He is among the co-authors of FrontierMath, released in November 2024, a benchmark of hundreds of original, exceptionally challenging mathematics problems crafted and vetted by expert mathematicians. At release, state-of-the-art AI models solve under 2% of its problems.

6. January–March 2025

   AGI, wages and the GATE model

   His January 2025 Epoch AI essay makes the case that AGI could drive human wages below subsistence level. In March 2025, he co-authors GATE, an integrated assessment model that simulates the economic effects of AI automation, and he and Ege Erdil examine their disagreements over AGI timelines on Epoch's podcast.

7. April–June 2025

   Co-founding Mechanize

   In April 2025, he leaves Epoch AI and launches Mechanize with Tamay Besiroglu and Ege Erdil; the startup develops virtual work environments, benchmarks and training data to enable the full automation of the economy. In June 2025, he estimates that AI agents able to take over computer work could be 10 to 20 years away.

8. April 2026

   Raising $9.1 million for Mechanize

   In April 2026, he, Besiroglu and Erdil announce that Mechanize has raised $9.1 million at a $500 million post-money valuation from Marco Mascorro, Adam D'Angelo and Devendra Chaplot.