Noah Ziems
@NoahZiems
PhD student @NotreDame advised by @Meng_CS. @DSPyOSS creator of Arbor
I've been working incredibly hard with @lateinteraction and the DSPy team on optimizing agentic workflows like DSPy programs. Excited to hear what everyone thinks! More announcements coming... 👀
So many things in the run-up to DSPy 3. Here's a first, EXPERIMENTAL one: 🚨We're releasing dspy.GRPO, an online RL optimizer for DSPy programs Your DSPy code as-is can be dspy.GRPO'ed. Yes, even compound multi-module programs. Led by @NoahZiems @LakshyAAAgrawal @dilarafsoylu.
Every time I go back to using Linux as my primary workhorse, it feels great at the beginning. But after a few months it feels like I’m maintaining the system rather than using it for what I want to do. I really hope DHH’s push here for a more opinionated Linux setup succeeds
I feel the same way about Linux now as I did Ruby in 2003: I have a moral obligation to present its beauty in the most accessible way possible. There are millions of developers who'd consider Linux if the whole package is presented with care, taste, and enthusiasm.
Can't make it to #ACL2025 this year, but for people interested in RL for factuality and textual decomposition, please check out our paper! TL;DR: We found a mismatch between the decomposition policy and LLM verifier, and propose a dynamic training paradigm to bridge the gap.
Big merge coming to Arbor soon! Lots of wonderful things coming together all at once
One of the best ways to get involved in research is to just start off by engineering something useful! If you keep your eyes open, you'll come across tons of new and interesting research directions that no one has thought of before
.@Sergeykarayev asks: What are some success stories with DSPy? Just for fun, we linked to TWENTY random success stories below in twenty minutes. We've seen hundreds more over the past 3 years, hence the 1M/mo downloads. Dozens of them are public, just search!
There are dozens and dozens. We post them literally all the time. Just scroll @DSPyOSS. Will collect a few examples for you below! There's no such thing as the DSPy team really. Almost everyone you see on here are regular users who love the framework.
You too can get a 32% improvement by simply switching to DSPy! And you’ll always be on top of the latest optimization strategies!
Had migrated to dspy.SIMBA from langchain implementation same demo or training data but max steps being 20 ( script runs in batch mode and near real time response not needed ) evaluate jumped by 32 percent ( avg metric per 1000 prompts ).
Grateful for the opportunity to lead a project of this scale and to work with some of the sharpest minds in this domain. Shoutout to @DSPyOSS and their incredible work in prompt optimization. Try out our new #AmazonNova prompt optimizer!
Introducing #AmazonNova prompt optimizer [Preview] 🤖🔄⚡ Our new open-source framework automates prompt optimization for Amazon Nova models, turning manual work into efficient, data-driven workflows. #AWS #generativeAI 👉 go.aws/3IllQwW
If only they could somehow write the general structure of their instructions and anytime a new closed-source model comes out they could optimize these instructions in a few minutes with only 3 lines of code🤔 Any ideas @DSPyOSS?
FYI cursor instructions are fine tuned for Claude only so other models are naturally lobotomized
I gave a talk on context engineering with DSPy! Come check out the full talk here: youtube.com/watch?v=1I9PoX…
Prompting isn’t engineering, argues @isaacbmiller1. “Context engineering is how we actually build reliable AI systems.” Learn how @DSPyOSS turns prompts into programs.
📢 New paper alert 📢 We introduce MobileGUI-RL, an RL framework advancing mobile GUI agents through trajectory-based rollouts and rewards in 𝗼𝗻𝗹𝗶𝗻𝗲 environments. With RL, Qwen 2.5-VL achieves 44.8% Success on Android World! ✨ Checkout paper at: arxiv.org/abs/2507.05720
Another wonderful weekend of working on DSPy + Arbor! So many cool things to look into!
You heard it here first folks. CEO of Shopify uses DSPy. Tobi, we have some really great stuff coming down the pipeline
DSPy is my context engineering tool of choice
# on technical accessibility One interesting observation I think back to often: - when I first published the micrograd repo, it got some traction on GitHub but then somewhat stagnated and it didn't seem that people cared much. - then I made the video building it from scratch,…