Will models be able to do the work of an AI researcher/engineer before 2027?

Question

In "Situational Awareness: The Decade Ahead", Leopold Aschenbrenner states:

I make the following claim: it is strikingly plausible that by 2027, models will be able to do the work of an AI researcher/engineer. That doesn’t require believing in sci-fi; it just requires believing in straight lines on a graph.

I will take "the work of an AI research/engineer" to mean all of the job duties of those roles, as we understand them today in 2024, but subject to small adjustments for the realities of 2027. Implicit in the claim behind this market is the idea that the world of 2027 could be strange enough that the claim no longer makes sense. I will use my best judgement. Adjust accordingly.

This is one of a series of markets on claims made in Leopold Aschenbrenner's Situational Awareness report(s):

@/Tossup/lasa-will-models-be-able-to-do-the

@/Tossup/lasa-100b-ai-training-cluster-by-th

@/Tossup/lasa-1t-ai-training-cluster-by-the

@/Tossup/lasa-by-the-end-of-2027-will-anyone

@/Tossup/lasa-by-the-end-of-2027-will-there

@/Tossup/lasa-will-a-tech-company-buy-an-alu

@/Tossup/lasa-manhattan-project-for-agi-befo

@/Tossup/lasa-1gw-ai-training-run-before-202

@/Tossup/lasa-10gw-training-run-before-2029

@/Tossup/lasa-100gw-ai-training-run-before-2

Other markets about Leopold's predictions:

@/ZviMowshowitz/will-we-develop-leopolds-dropin-rem

Manifold Markets · Answer

Unlikely — Manifold Markets prediction market estimates a 13% chance (48 traders, as of Mar 16, 2026).