Shengran Hu (@shengranhu) / X

Shengran Hu

307 posts

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

Joined February 2019

Pinned
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
23K
Shengran Hu
@shengranhu
Aug 19, 2024
Thrilled to introduce Automated Design of Agentic Systems (ADAS), a step toward full automation in agentic systems development from the bottom up in a self-referential manner. Delighted to work with @cong_ml and @jeffclune on this! Website and code:
Jeff Clune
@jeffclune
Aug 19, 2024
Can AI agents design better AI agents? We describe a newly forming research area "Automated Design of Agentic Systems" (ADAS) that aims to automatically design powerful agents & a new method where a meta agent invents new agents by programming them in code. Led by @shengranhu 🧵
shengranhu.com
ADAS
Project page for Automated Design of Agentic Systems.
12K
Shengran Hu
@shengranhu
May 30, 2025
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
Sakana AI
@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,
15K
Shengran Hu
@shengranhu
Sep 11, 2024
Fascinating work on Visualization-of-Thought in LLMs/agents. Interestingly, agents with similar visual reasoning techniques also emerged from our Automated Design of Agentic System (ADAS) algorithm and became one of the best agents in math domains!
Carlos E. Perez
@IntuitMachine
Apr 21, 2024
1/n Beyond Words: Visualizing Thought to Unlock AI's Spatial Reasoning Imagine yourself dropped into a bustling city, armed with nothing but a cryptic set of directions and your own intuition. To reach your destination, you must decipher the instructions, visualize the route,
12K
Shengran Hu
@shengranhu
Dec 16, 2024
Extremely honored that our paper Automated Design of Agentic Systems (w/ @cong_ml and @jeffclune) received the Outstanding Paper Award at the #NeurIPS Open-World Agent workshop. Thanks very much to the organizers of the workshop!
21K
Shengran Hu
@shengranhu
Mar 12, 2025
Thrilled to share that a paper produced by The AI Scientist-v2 passed peer review at an ICLR workshop!
Sakana AI
@SakanaAILabs
Mar 12, 2025
The AI Scientist Generates its First Peer-Reviewed Scientific Publication We’re proud to announce that a paper produced by The AI Scientist-v2 passed the peer-review process at a workshop in ICLR, a top AI conference. Read more about this experiment → sakana.ai/ai-scientist-f…
4.4K
Shengran Hu
@shengranhu
Jan 23, 2025
Two of our papers were accepted to #ICLR2025! Thank you to my incredible collaborators @cong_ml and @jeffclune🥰 See you all in Singapore 🇸🇬🏖️ [1/2] Automated Design of Agentic Systems:
Jeff Clune
@jeffclune
Aug 19, 2024
Can AI agents design better AI agents? We describe a newly forming research area "Automated Design of Agentic Systems" (ADAS) that aims to automatically design powerful agents & a new method where a meta agent invents new agents by programming them in code. Led by @shengranhu 🧵
4.3K
Shengran Hu
@shengranhu
Feb 12, 2025
Thrilled to introduce ACD: AI that discovers its own capabilities & failure modes by endlessly generating novel tasks 🤖🔍 Absolute joy collaborating with @cong_ml & @jeffclune on this work!
Jeff Clune
@jeffclune
Feb 12, 2025
Introducing Automated Capability Discovery! ACD automatically identifies surprising new capabilities and failure modes in foundation models, via "self-exploration" (models exploring their own abilities). Led by @cong_ml & @shengranhu 🔬🤖🧠🔎 [1/9]
2.9K
Shengran Hu
@shengranhu
Dec 9, 2024
Our work Automated Design of Agentic Systems (w/ @cong_ml and @jeffclune) will have ✨two orals✨ @ #NeurIPS2024 workshops (LanGame Sat 10:20, OWA Sun 4:50). Please come visit us😃 I would also love to chat about open-endedness, LLM agents, etc. Let me know if you want to meet!
6.5K
Shengran Hu
@shengranhu
Nov 5, 2024
Replying to @cwolferesearch
Automated prompt optimization is awesome. But many components in agentic systems could be optimized too (workflow, tool use, memory, etc.) Our paper ADAS attempts to define agents in code and optimize all possible components.
shengranhu.com
ADAS
Project page for Automated Design of Agentic Systems.
2.4K
Shengran Hu
@shengranhu
Apr 23, 2025
I will be attending ICLR 2025 in Singapore 🇸🇬! Two main conference posters (w/ @cong_ml & @jeffclune): - Intelligent Go-Explore ⏰Thu 24, 3 pm - 5:30 pm 📍Hall 3 + Hall 2B #282 - Automated Design of Agentic Systems ⏰Sat 26, 3 pm - 5:30 pm 📍Hall 3 + Hall 2B #275 🧵
3K
Shengran Hu
@shengranhu
Jun 19, 2023
Excited to introduce Thought Cloning! We've open-sourced the code and dataset. Check more details at: shengranhu.com/ThoughtCloning/.
Jeff Clune
@jeffclune
Jun 2, 2023
Introducing Thought Cloning: AI agents learn to *think* & act like humans by imitating the thoughts & actions of humans thinking out loud while acting, enhancing performance, efficiency, generalization, AI Safety & Interpretability. Led by @shengranhu arxiv.org/abs/2306.00323 1/5
00:00
shengranhu.com
Thought Cloning
Project page for Thought Cloning: Learning to Think while Acting by Imitating Human Thinking.
5.1K
Shengran Hu
@shengranhu
Sep 3, 2024
It was a great pleasure to discuss our work on 'Automated Design of Agentic Systems (ADAS)' shengranhu.com/ADAS/ on the TWIML podcast!
The TWIML AI Podcast
@twimlai
Sep 3, 2024
Today we're joined by @shengranhu, a PhD student at @UBC_CS to discuss Automated Design of Agentic Systems (ADAS). We dig into the spectrum of agentic behaviors, the key ideas behind ADAS, and how it uses LLMs to design novel agent architectures in code. We explore its insights
00:00
shengranhu.com
ADAS
Project page for Automated Design of Agentic Systems.
4.1K
Shengran Hu
@shengranhu
May 11, 2025
Check out ADAS (shengranhu.com/ADAS/) and many recent works in this direction - not only learning system prompts, but also learning all components in agentic systems
Andrej Karpathy
@karpathy
May 11, 2025
We're missing (at least one) major paradigm for LLM learning. Not sure what to call it, possibly it has a name - system prompt learning? Pretraining is for knowledge. Finetuning (SL/RL) is for habitual behavior. Both of these involve a change in parameters but a lot of human
shengranhu.com
ADAS
Project page for Automated Design of Agentic Systems.
1.1K