Twitxy Saved

Shengran Hu

@shengranhu

Founding Member @Recursive_SI. PhD @UBC_CS. Building self-improving AI for knowledge discovery. ADAS | DGM | The AI Scientist
Vancouver, British Columbia shengranhu.com Joined February 2019
554 Following 1.9K Followers 327 Posts

Latest posts 5

Pinned post

Shengran Hu @shengranhu
Can AI systems improve themselves recursively—agents that enhance their performance while advancing further self-improvement? Today, we take a step toward realizing this vision.

Introducing Darwin Gödel Machine: AI that improves itself by rewriting its own code.

We build on the concept of the Gödel Machine, a hypothetical recursively self-improving computer program that solves problems in an optimal way, but drop its impractical assumption of mathematically proving the improvement. Instead, we leverage open-ended algorithms inspired by Darwinian evolution to search for empirically validated performance improvements.

Advancing beyond the Automated Design of Agentic Systems (ADAS), where a fixed meta agent creates new agents with, e.g., enhanced tools and workflows, this next evolution enables agents to continuously refine themselves for self-referential improvements.

We prioritize safety throughout the development of this project; please see our blog and paper for detailed discussions. Provided that the safety concerns are carefully navigated, DGM could pave the way for open-ended self-referential AI improvements, continually rewriting or retraining itself in pursuit of greater capabilities aligned with human values.

Co-led by @jennyzhangzt and me, in collaboration with @cong_ml, @RobertTLange, and @jeffclune. It is an absolute joy to work with my incredible collaborators!
Sakana AI Verified account @SakanaAILabs May 30, 2025
Introducing The Darwin Gödel Machine: AI that improves itself by rewriting its own code

sakana.ai/dgm

The Darwin Gödel Machine (DGM) is a self-improving agent that can modify its own code. Inspired by evolution, we maintain an expanding lineage of agent variants, allowing for open-ended exploration of the vast design space of such “self-improving” agents.

Modern agentic systems, while powerful, remain static—once deployed, their intelligence remains fixed. We believe continuous self-improvement is key to the development of stronger AI capabilities. Our Darwin Gödel Machine is built from the ground up to enable AI systems that can learn and evolve their own capabilities over time, just as humans do.

On SWE-bench, DGM automatically improved its performance from 20.0% to 50.0%. Similarly, on Polyglot, the DGM increased its success rate from an initial 14.2% to 30.7%, significantly outperforming representative hand-designed agents.

Learn more about our approach in our technical report: arxiv.org/abs/2505.22954

This work was done in collaboration with Jeff Clune (@jeffclune)’s lab at UBC, and led by his PhD students Jenny Zhang (@jennyzhangzt) and Shengran Hu (@shengranhu), together with Cong Lu (@cong_ml) and Robert Lange (@RobertTLange).

Code: github.com/jennyzzt/dgm
View quoted post on X
14 replies 18 reposts 98 likes 15K views View on X
Shengran Hu @shengranhu
Honored to see our work on ADAS, DGM, and The AI Scientist show up in @lilianweng's blog — been a big fan for years!
Lilian Weng Verified account @lilianweng Jul 7
new post on harness engineering for AI self-improvement: lilianweng.github.io/posts/2026-07-…

It is hard to forecast how much the future of RSI will rely on harnesses. Likely harness engineering will evolve in the direction of self-improvement and enable auto-research, and, in turn, smarter models keeps harnesses simple.

Even when many harness improvement get eventually internalized into core model, the need to specify goals and context will not disappear.
View quoted post on X
1 replies 0 reposts 18 likes 1.9K views View on X
Shengran Hu @shengranhu
(also quietly hit 1,000 citations today :))
Shengran Hu @shengranhu May 13
Some updates: I've joined Recursive as a member of the founding team.

My core curiosity about the world centers on how complex patterns and knowledge emerge from the two open-ended processes we know: natural and cultural evolution. I've been lucky to explore this during my PhD through works like ADAS, Darwin Gödel Machine, and The AI Scientist. Excited to keep chasing this thread with the incredible team!
View quoted post on X