Lu Wang

🚀 At Instacart, we just launched PIXEL — our in-house generative AI platform for product imagery. From DreamBooth fine-tuning to automated quality review with VLMs, PIXEL helps teams generate high-quality food images 10x faster — boosting conversions and reducing manual effort. Proud of the cross-functional AI work here! #GenerativeAI #ComputerVision #Instacart #ML

28

** Big update on the roadmap for Meta's MTIA roadmap (our Meta-designed AI chips) ** We just shared today a big update on the MTIA (Meta Training & Inference Accelerator) roadmap. The main takeaway is we're accelerating the whole set of programs to get rapid iterations on chips (given how fast the model needs are evolving). 👉 Overall blog is here: https://lnkd.in/gaHuxUfg For the networking aficionados, you can see a whole lot of networking that we've been working on inside MTIA - we're excited to finally be able talk about it! 👉 The scale-up and scale-out networks we've already talked about. It's sort of buried in there, but we are already running distributed training clusters with MTIA 300 for core ranking & recommendation (even as we focus future generations on Inference). 👉 The communication library (HCCL) that is purpose-built to take advantage of networking-specific capabilities we've built into the chips to offload the main compute. 👉 The in-built support for fused compute/communication kernels to minimize latency (key for both training and inference workloads). Super-proud of the whole MTIA team!

143

4 Comments

People are leveraging AI to build semantic layers. People are leveraging semantic layers to use AI. That just sounds too funny to me.

4

3 Comments

Truly transformative. Within the next 6 months coding will be fully commoditized. Only design, distribution, and hardware will be the moat. It is time to lean in #AI #FutureofWork #OhioTech

1

1 Comment

Third piece in the AAD series is up. This one is about acceptance criteria (AC) and why treating them as a formal contract changes what you can diagnose when autonomous development breaks.  AC serve as the bridge between the spec and the environment in which the spec is compiled into a product. This is implicit in the first article, but after marinating on this for a few weeks, I felt like I should make it explicit.                                                                                                    The short version: most teams write AC the way you'd write them for a human. "Handle edge cases gracefully." "Performance should be acceptable under normal load." In AAD, those are  specification failures. An agent will do something when it reads that. It might even do something reasonable. But you've handed control over what "reasonable" looks like to the LLM, and you'll only find out whether you agree after you've burned the iteration cycle. In my experience, the first pass is ok, but it can quickly diverge and it takes a lot of time to get back to where you intended to go in the first place. One thing I keep coming back to is the diagnostic value of keeping the spec and the AC separate. My currently line of thinking leads me to this framing: The spec describes intent. The AC defines done. You need both. The Spec is the what and the why. The AC are the falsifiable tests that the intention was met. Keeping them separate creates a framework so they each get the attention they deserve as part of this new world. Building on the notion of falsifiability, there's also a section on the right mindset for writing AC. Just like with the best scientist/experimentalist, you should be writing AC from a lens that lets you confirm intention was not met, not that intention was met. It's a subtle but meaningful nuance. If you assume the agent is going to get it correct, you're biased to see the final product and assume it's correct. When you train yourself to focus on the ways the agent might get it wrong, you open up a whole new world on writing great AC. I think this applies to traditional development as well as AAD. Claude made a fancy doc for this one! But also, the full read is in the comments.

15

1 Comment

A thoughtful piece by our own Brian de Luna on Applied Science in marketplaces, and the unique journey we’ve taken at Faire.

18

In the era of LLMs, delivering reliable, robust, and trustworthy conversational systems means more than picking a model — it means embedding rigorous evaluation throughout the entire development lifecycle. 🔍 What we cover in the blueprint: - From dataset curation (public & internal data) to synthetic query generation - To designing metrics and rubrics that go beyond BLEU / ROUGE - To building an evaluation platform to manage versioning, experiments, and traceability - To layered gates and automated checks at pull-request, staging, and production levels - To continuous improvement loops, mining real-world failures to guide the next iteration 💡 Key insight: Evaluation is not a “post-hoc step” — it must be treated with the same rigor as production code. Every prompt tweak, retriever update, or model swap must pass through reproducible, automated checks. Tiny changes can cascade, and without guardrails, regressions bleed into production. If you’re building conversational AI — whether retrieval-augmented generation systems, QA assistants, or multimodal agents — we hope this blueprint helps you construct a robust evaluation-first pipeline that scales. Check out the full write-up here: https://lnkd.in/eAkpm2yg We’d love to hear your thoughts or experiences — what evaluation practices have worked (or failed) for you? #AI #MachineLearning #ConversationalAI #LLMs #Evaluation #MLOps

51

3 Comments

Among other things, my team has been focusing on AI research for agents. We just blogged about a recent project: the Test-Time Diffusion Deep Researcher (TTD-DR). This framework uses a Deep Research agent to draft and revise its own drafts using high-quality retrieved information. This approach achieves new state-of-the-art results in writing long-form research reports and completing complex reasoning tasks. The work achieves better results than any other deep research agent we compared to. It also shows that self-evaluation and self-improvement are must-have components of a successful agent. https://lnkd.in/gBZpVMQy

297

8 Comments

Could non-technical staff build 9x AI workflows in one day? A week or so ago, I ran a one-day hackathon inviting engineers and colleagues from onboarding, RevOps, support and operations. The winning team's fraud detector identified a bizarre pattern; a customer buying a chicken carcass every few minutes for two days straight... 🐓 Over 20 participants formed 9 cross-functional teams and, by day’s end, each pitched an AI workflow: - CS and knowledge-base chatbot, - underperforming merchant alerts, - renewal-email sentiment analysis, - fraud detection 🥇(winning team!), - sales-opportunity optimiser, - loyalty-socials content generator, - optimised outbound workflow, - CI deprecation solver, - lead qualification and more... 𝗛𝗲𝗿𝗲'𝘀 𝗺𝘆 𝗽𝗿𝗮𝗰𝘁𝗶𝗰𝗮𝗹 𝗮𝗱𝘃𝗶𝗰𝗲.. Within the next week, invite one non-technical colleague to co-create a simple Agentic workflow. Dedicate four hours to tackle a routine process, such as handling support tickets or generating renewal reminders, using a no-code interface that connects to your data (via an MCP server) such as N8N or Zapier. Focus on a clear pain point, prototype rapidly and iterate based on immediate feedback. What cross-functional AI workflow will your team prototype next? Let me know in the comments if you get stuck! #AIWorkflows #CrossFunctionalAI #Hackathon #AgenticAI #DataInnovation #DiverseTeams

23

1 Comment

#1 - Toss a ball to a friend that is running at an angle and have them catch it. Not that hard for humans in general. Very hard for AI. #2 - Document and explain the math behind #1, with variants for wind speed, elevation, etc. Very hard for humans, relatively trivial for AI.  There's a lot of opportunity at the jagged frontier.

8

🧠 While LLMs continue to grab the spotlight, Small Language Models (SLMs) are quietly reshaping how we think about efficiency in AI. I recently explored how SLMs can run entirely on CPUs, unlocking new possibilities for low-cost, privacy-friendly, and scalable inference, ideal for teams experimenting with AI or deploying it at the edge. In my latest piece for Towards AI publication, I cover: 💡 Why CPUs are back in the AI conversation: smarter models, quantization, and optimized runtimes like llama.cpp ⚙️ SLMs vs LLMs: how a hybrid strategy can balance capability and cost 🧩 CPU Inference Tech Stack: GGUF, Ollama, and other tools making CPU inference practical 🧠 Hands-on section: building a translation app using llama.cpp, Streamlit, and AWS EC2, fully CPU-based The best part? You can try it locally - no GPU required. It’s been exciting to see the response so far, the article has already crossed 200+ likes on Medium! 📖 Read my full blog post here: 👉 https://lnkd.in/gby-EBmJ #SLM #llamacpp #cpu #BeginnerFriendly #AWS

7