Sharon Zhou (@realsharonzhou) 's Twitter Profile
Sharon Zhou

@realsharonzhou

Building the future of LLMs | Cofounder & CEO, @LaminiAI | Prev: CS Faculty & PhD @Stanford. Product @Google. @Harvard | @MIT 35 under 35. Angel investor.

ID: 796109256000843776

linkhttp://lamini.ai calendar_today08-11-2016 21:56:52

1,1K Tweet

24,24K Followers

1 Following

Lamini (@laminiai) 's Twitter Profile Photo

Have you seen our Classifier Agent Toolkit 😺 demo yet? Learn how to use our SDK to build a highly accurate Classifier Agent for a customer service chatbot. The agent categorizes customer interactions by intent so it can respond appropriately. You can run multiple evaluations

Sharon Zhou (@realsharonzhou) 's Twitter Profile Photo

🚀 Introducing Memory RAG: Using embed-time compute for 90%+ RAG accuracy on mini LLMs — for a future of specialized, capable, (cheap) mini-agents! 🤖 The core insight: Most RAG systems waste compute during inference. We moved it to embedding generation instead. 💡 The big

🚀 Introducing Memory RAG: Using embed-time compute for 90%+ RAG accuracy on mini LLMs — for a future of specialized, capable, (cheap) mini-agents! 🤖 

The core insight: Most RAG systems waste compute during inference. We moved it to embedding generation instead. 💡 

The big
Databricks (@databricks) 's Twitter Profile Photo

General LLMs are pretty good at everything, but perfect at nothing. Lamini CEO and co-founder Sharon Zhou joined attendees at #DataAISummit 2025 to explore building compound AI systems by combining SLMs with highly accurate mini-agents — demonstrating techniques like

General LLMs are pretty good at everything, but perfect at nothing.

<a href="/LaminiAI/">Lamini</a> CEO and co-founder <a href="/realSharonZhou/">Sharon Zhou</a> joined attendees at #DataAISummit 2025 to explore building compound AI systems by combining SLMs with highly accurate mini-agents — demonstrating techniques like
Sharon Zhou (@realsharonzhou) 's Twitter Profile Photo

Excited to share big news! 🎉 I'm joining Lisa Su at AMD to work on what I love most: AI research & teaching Dream: Everyone becomes GPU-rich, scaling laws hit their asymptotic limits, and we democratize those sweet matmuls Several intense, cute Laminati from Lamini are

Excited to share big news! 🎉 I'm joining <a href="/LisaSu/">Lisa Su</a> at <a href="/AMD/">AMD</a> to work on what I love most: AI research &amp; teaching

Dream: Everyone becomes GPU-rich, scaling laws hit their asymptotic limits, and we democratize those sweet matmuls

Several intense, cute Laminati from <a href="/LaminiAI/">Lamini</a> are
Sharon Zhou (@realsharonzhou) 's Twitter Profile Photo

Excited to moderate a fireside chat between two incredible AI leaders, who happen to be my current and former bosses: Dr. Lisa Su and Dr. Andrew Ng ! Come toast some GPU-shaped marshmallows with us! Hope I don’t get roasted 🔥 — in the Luminary Room at #AdvancingAI 📸:

Excited to moderate a fireside chat between two incredible AI leaders, who happen to be my current and former bosses: Dr. <a href="/LisaSu/">Lisa Su</a> and Dr. <a href="/AndrewYNg/">Andrew Ng</a> !

Come toast some GPU-shaped marshmallows with us! Hope I don’t get roasted 🔥

— in the Luminary Room at #AdvancingAI 

📸:
AMD (@amd) 's Twitter Profile Photo

Open source isn’t just a philosophy, it’s a force multiplier for AI progress. Sharon Zhou, VP of AI at AMD, shares why open ecosystems are critical to unlocking the full potential of generative AI. From enabling community-driven innovation to fueling a virtuous cycle of

AI at AMD (@aiatamd) 's Twitter Profile Photo

What if AI could improve itself? Sharon Zhou, VP of AI at AMD, shares her vision for self-evolving systems, where models not only learn from data but also generate and refine it to get smarter over time. #AdvancingAI

Sharon Zhou (@realsharonzhou) 's Twitter Profile Photo

The story of hybrid architectures is honestly fascinating! I've been diving deep into why Transformers became the default choice, and looking at new model architectures. It's not because "attention is all you need" (though catchy!) It's because they exploited GPU parallelism so

The story of hybrid architectures is honestly fascinating! I've been diving deep into why Transformers became the default choice, and looking at new model architectures. It's not because "attention is all you need" (though catchy!) 

It's because they exploited GPU parallelism so
Sharon Zhou (@realsharonzhou) 's Twitter Profile Photo

Riffing on this. One issue is RL in post-training leans more towards redistributing probabilities over discovering new modes. What if LLMs could learn from “vicarious rewards”? As a person, I spend a lot of bandwidth observing social cues of other humans, their successes and

Sharon Zhou (@realsharonzhou) 's Twitter Profile Photo

Happy weekend :) Sharing a quick fun video explaining the basics of Vision-Language Models, VLMs! VLMs are used for things like chatting with an image or identifying text in an image - and can handle super blurry images and over time to learn to identify text and features better