AI-Driven Research Systems
@ai4research_ucb
Followers
302
Following
29
Media
4
Statuses
19
Joined October 2025
Thought-provoking efforts and article that raise the discussions on future systems research with AI -- if AI could take over algorithm discovery, what roles should human researchers play and what are the best practices of using AI in systems research? The paper title is
🚀 Excited to release our new paper: “Barbarians at the Gate: How AI is Upending Systems Research” We show how AI-Driven Research for Systems (ADRS) can rediscover or outperform human-designed algorithms across cloud scheduling, MoE expert load balancing, LLM-SQL optimization,
1
2
29
It’s truly impressive to be part of the evolution of the research paradigm itself! The evolved EPLB algorithm will soon be integrated into vLLM - give it a try then!
🚀 Excited to release our new paper: “Barbarians at the Gate: How AI is Upending Systems Research” We show how AI-Driven Research for Systems (ADRS) can rediscover or outperform human-designed algorithms across cloud scheduling, MoE expert load balancing, LLM-SQL optimization,
0
3
7
Excited to share our new paper on AI-Driven Research for Systems. We show that AI can autonomously generate and verify novel solutions for classic systems performance problems, matching or exceeding human designs. A glimpse into how AI might transform not only systems, but the
🚀 Excited to release our new paper: “Barbarians at the Gate: How AI is Upending Systems Research” We show how AI-Driven Research for Systems (ADRS) can rediscover or outperform human-designed algorithms across cloud scheduling, MoE expert load balancing, LLM-SQL optimization,
5
32
173
We've been getting some great results with AI for systems optimization at Berkeley! This is a way of using AI that I think is under-explored: use LLMs as a prior to search for a solution given a verifiable metric. It could probably be applied to more industry problems.
🚀 Excited to release our new paper: “Barbarians at the Gate: How AI is Upending Systems Research” We show how AI-Driven Research for Systems (ADRS) can rediscover or outperform human-designed algorithms across cloud scheduling, MoE expert load balancing, LLM-SQL optimization,
2
19
92
@ai4research_ucb Excited to share our new paper on AI-Driven Research for Systems. We show that AI can autonomously generate and verify novel solutions for classic systems performance problems, matching or exceeding human designs. A glimpse into how AI might transform not only systems, but the
0
6
35
I’ve spent much of my PhD designing algorithms to optimize system performance. Now, our new research shows AI can discover solutions that are 5x faster & 26% cheaper—in hours. It's a profound shift in how we do research. Welcome to the age of AI-Driven Research.
🚀 Excited to release our new paper: “Barbarians at the Gate: How AI is Upending Systems Research” We show how AI-Driven Research for Systems (ADRS) can rediscover or outperform human-designed algorithms across cloud scheduling, MoE expert load balancing, LLM-SQL optimization,
1
4
24
AI is going to replace researchers? 🙀 AI is going to replace PhD students? 🙀 AI is going to take all of our jobs? 🙀 NO! But… 🚨AI is upending systems research 🚨 We show that by leveraging AI-driven research systems (ADRS), we can drastically accelerate the algorithm
🚀 Excited to release our new paper: “Barbarians at the Gate: How AI is Upending Systems Research” We show how AI-Driven Research for Systems (ADRS) can rediscover or outperform human-designed algorithms across cloud scheduling, MoE expert load balancing, LLM-SQL optimization,
2
10
71
Over the past few years, I've spent countless ⏰ tuning schedulers, cache policies, and placement algorithms by hand. 🤯 Turns out that AI can now do that for us — and sometimes even better! Thrilled to share our new paper below, where we show how AI can rediscover or
🚀 Excited to release our new paper: “Barbarians at the Gate: How AI is Upending Systems Research” We show how AI-Driven Research for Systems (ADRS) can rediscover or outperform human-designed algorithms across cloud scheduling, MoE expert load balancing, LLM-SQL optimization,
1
7
28
(9/N) This is a joint collaboration from Sky Computing lab with @audreyccheng, @LynnLiu41887950, @melissapan, @andylizf, Bowen Wang, @AlexKrentsel, @tian_xia_, @mertcemri, @Jongseok_Park_, @randwalk0, Jeff Chen, @LakshyAAAgrawal, @Apd10Desai, @Jiarong_Xing, @koushik77,
2
4
13
(8/N) 👋 Join us! This is an early effort, and we hope to iterate on the APIs and architecture of ADRS with help and feedback from the community. Please leave your comments, and don’t hesitate to reach out. Follow us for future research, blog posts, and exciting updates!
1
3
13
(7/N) 📢 Barbarians at the Gate is a call to action to the systems research community As AI begins mastering algorithm discovery, researchers’ roles will shift — from designing solutions to focusing on problem specifications.
1
1
9
(6/N) 🧠What we learned: evolving algorithms with AI is powerful but tricky. 1. Start worse to get better – Overly-optimized initial code restricts AI creativity. A simple baseline can lead to bigger breakthroughs. 2. Prevent overfitting — Use test traces that cover the real
1
1
9
(5/N) 🤔 Why systems? Because system performance problems are measurable, easy to verify, and cheap to test. -- Throughput, latency, or cost – all have objective evaluators. -- Simulators used frequently in systems domain make it possible to explore thousands of algorithm
1
2
7
(4/N) 💡AI-Driven Research for Systems (ADRS) changes this: it automates algorithm design — letting researchers focus on what to optimize, not how. Given a problem, it iteratively generates, tests, and evolves solutions automatically. Simulators or real testbeds verify
1
1
9
(3/N) Traditionally, systems research takes months — humans design, implement, and test schedulers, allocators, or protocols by hand. 🧑💻 PhD students spend ~40% of their time of their time just iterating on designs and experiments.
1
1
10
(2/N) Across 11 case studies — from cloud scheduling to load balancing to LLM inference — ADRS rediscovered or outperformed published systems results: 📊 Up to 5× faster or 26% cheaper Example: • ☁️Multi-region job scheduling → 26% lower cost • 🎓 MoE load balancing → 5×
1
2
12
🚀 Excited to release our new paper: “Barbarians at the Gate: How AI is Upending Systems Research” We show how AI-Driven Research for Systems (ADRS) can rediscover or outperform human-designed algorithms across cloud scheduling, MoE expert load balancing, LLM-SQL optimization,
8
32
130