ClaimArticle

Our automated AI research system's new state-of-the-art results are an early sign it can push the frontier on AI training and infrastructure tasks, at least when goals are well-defined, measurable, and fast to evaluate.

Startup Recursive reported new state-of-the-art results on small-model training and GPU kernel optimization benchmarks using its automated research-loop system, framing it as an early proof point for recursive self-improvement. ✦ AI generated

Recursive · Import AI · 2026-06-22 · original ↗

Share card ↓Export .md Export .txt

These results are an early sign that our system can push the frontier on AI training and infrastructure tasks, especially when the goal is well-defined, measurable, and quick enough to evaluate many times

Read full article ↗excerpt · fair-use quotation

▦ Recursive ✳ NanoGPT Speedrun ✳ SOL-ExecBench

Around this claim

Evidence · 2

Intology's Locus harness under-elicits today's AI systems' ability to automate AI R&D, and the human baseline on PostTrainBench v1.1 (51.1%) will likely be exceeded before the end of 2026.Jack Clark · Import AI · conf 65%While AI systems struggle with original research creativity, OpenAI has solved ten verified open problems in mathematics and theoretical computer science using an internal version of Astra, signaling AI's arrival in domains where verifiability replaces subjective taste.Jack Clark (Import AI), quoting OpenAI and Henry Yuen · Import AI · conf 60%

In practice · 1

The future of research is a shepherd-ship paradigm where humans steer distributed, multi-threaded AI systems that run experiments autonomously.Robert Lange · Machine Learning Street Talk · conf 85%

This moment responds to

supports → Artificial superintelligence may arrive within a few years, and it is unclear whether alignment research is on track to be ready in that same timeframe, since current empirical programs at AI labs are unlikely to deliver a priori confidence that training an ASI will go well.Sequent · Import AI