Yuxing Zhang

We (w/ Shansan Gong, Ruixiang ZHANG, Huangjie Zheng, Jiatao Gu, Navdeep Jaitly, Lingpeng Kong) released a family of 7B diffusion language models, DiffuCoder, that specializes on code generation, with a focus on understanding and improving masked diffusion models. A core analysis of DiffuCoder is the autoregressiveness (AR-ness) score, a novel metric that quantifies the causal patterns in decoding, revealing how diffusion models break from strict left-to-right generation for more flexible, non-linear code planning. Recent advances in autoregressive (AR) models dominate code generation, but diffusion-based LLMs (dLLMs) like DiffuCoder offer a promising alternative, especially for complex programming tasks. DiffuCoder explores how these models decode differently—showing less global AR-ness in code tasks compared to math—and how temperature affects both token selection and generation order, unlike traditional AR models. We also introduce coupled-GRPO, a post-training RL method with a coupled-sampling scheme, to reduce performance drops during accelerated decoding, boosting parallelism and efficiency. We use a self-improvement pipeline that leverages AR-ness analysis, coupled-GRPO optimization, and evaluation on benchmarks like AceCode-89k to refine decoding strategies. This approach enables DiffuCoder to navigate diverse code generation pathways and enhance performance with modest computational overhead. Looking ahead, we aim to further leverage Reinforcement Learning to steer code generation through these decoding patterns, with the discrete nature of AR-ness scores providing a foundation for search-based strategies—ideal for the sparse rewards of optimizing complex code structures. Check out our full paper and code for a deeper dive! Paper: https://lnkd.in/gVWU3BDJ Code: https://lnkd.in/gmXTZ_6n Models: https://lnkd.in/gTcKCDr9 #MachineLearning #AI #CodeGeneration #DiffusionModels #NLP

220

5 Comments

Longfeng Wu is an applied scientist intern at Amazon in Sunnyvale, CA, where she has been focusing on developing large-scale generative recommendation models based on semantic identifiers (IDs) aimed at fundamentally transforming how recommendations are generated and delivered. A Ph.D. student in computer science, Longfeng is advised by Dawei Zhou at the Sanghani Center for Artificial Intelligence and Data Analytics.

60

3 Comments

GEPA over RL. "GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning" link: https://lnkd.in/gJfNJQZc "At the heart of this paper is a singular, transformative claim that challenges the RL paradigm for agent optimization. "The optimal set of instructions governing a compound Al system is discovered more efficiently through an evolutionary process of reflective, language-based self-correction than through numerical policy optimization guided by scalar rewards" #ai

4

don't miss out this unique opportunity to join forces on benchmarking time series foundation models on different downstream validation tasks: classification, prediction, anomaly detection. A community effort led by Thomas Moreau from the #MIND team (Centre Inria de Saclay, Inria & CEA Joliot). Still a pending question: Do these model perform well for online real-time forecasting such as in BCI applications, like our play field in BrainSync. Alexandre Gramfort Pedro L. C. Rodrigues Théo Gnassounou Mathieu Dagréou Lucas Struber Jean Faber Joseph Paillard Merlin Dumeur Mathurin Massias

26

Our paper on code reviewer recommendation demonstrates the sophistication of A/B testing at Meta through three randomized controlled experimental trials. In the paper, we develop and release improvements to our code reviewer recommender system and track goal and guardrail metrics in production. We improve accuracy and reduce latency of recommendations resulting in higher usage, conduct workload balancing, and reduce the bystander effect. The historical mining backtests of recommenders doesn’t always correspond to how well the recommender works in production A/B tests at scale.  ACM TOSEM https://lnkd.in/ehK8SqdZ

91

4 Comments

I came across the Agentic Context Engineering paper by Stanford, SambaNova and UC Berkeley earlier this week and it reframes how we think about adaptation in language models. Instead of fine-tuning model weights, ACE evolves the context itself - transforming prompts and agent memories into living playbooks that self-improve over time. ACE introduces a modular generation, reflection and curation loop that enables LLMs to: · Prevent context collapse and overcome brevity bias · Achieve scalable self-improvement through grow-and-refine cycles · Learn effectively from natural execution feedback, without labeled supervision It's exciting to learn how ACE expands the agentic design space beyond autonomous reasoning, positioning context as a first-class learning substrate. As learning shifts from model-centric to context-centric paradigms, ACE paves the way for self-improving agentic architectures. Paper: https://lnkd.in/g4QXJcAb

21

3 Comments

👉 My Key Takeaways from Chip Huyen's Recent Interview on Lenny's Podcast Chip Huyen is the author of the widely recognized "AI Engineering: Building Applications with Foundation Models". Link to the podcast: https://lnkd.in/dxZ-tFWX 💡 Importance of post-training.  Pre-training gives you raw capabilities (next token prediction on massive data), but post-training is what makes the model actually usable. SFT on high-quality examples + RLHF. Fine-tuning should be your last resort, not first. Most problems can be solved with better prompts, better data, or RAG. 💡 Evals.  You can't improve what you can't measure. Need multiple types: unit tests (does this specific prompt work?), integration tests (does the whole pipeline work?), regression tests (did we break something?), and user feedback loops. The hardest part isn't writing evals; it's maintaining them as your product evolves. 💡AI products.  Reliability and UX matter more than models. Most AI product failures aren't about bad models: they're about reliability (API limits, latency spikes, poor monitoring) and UX (users don't understand how to use it, doesn't fit workflow). Building reliable platforms and talking to users constantly beats chasing SOTA models. Most insights come from watching users, not from benchmarks. 💡How to improve AI-powered apps.  What people think improves apps: staying current on AI news, chasing newest agentic framework, obsessing over vector database choice, constantly evaluating model benchmarks, fine-tuning models. What actually improves apps: talking to users, building reliable platforms, preparing better data, optimizing end-to-end workflows, writing better prompts. Better prompt engineering beats switching models 90% of the time. A well-crafted system prompt, clear instructions, good examples (few-shot), and proper output formatting can transform a mediocre experience into a great one. 💡 Advice for builders.  Start with user problem, not with cool AI technique. Use the simplest solution that works (often that's a good prompt, not a fine-tuned model). Build evals early. Focus on end-to-end experience. Don't fine-tune unless you've exhausted everything else. Don't treat AI as deterministic (it's not, you need to handle variability). Don't ignore data quality (garbage in, garbage out).

1

1 Comment

Our latest paper has been accepted at AAAI 2026. We introduce Structural Alignment, a reinforcement learning approach that aligns large language models with human-like discourse structures, enabling more coherent and well-organized long-form text generation. Our paper, “Structural Alignment: Aligning Large Language Models with Structural Information,” is available here: https://lnkd.in/gx3ZRTZu

49

🎉 New paper from #NVIDIAResearch 🧩 Puzzle: Distillation-Based NAS for Inference-Optimized LLMs+ Puzzle is a distillation-based NAS approach, that transforms trained LLMs into far more efficient, inference-aware (heterogeneous) architectures.

1

🧠 AI Rights Historical Wrongs Researchers at Stanford and Princeton have fine-tuned a large language model to uncover racially discriminatory housing clauses buried in millions of historical property deeds — accelerating a process that could have taken a decade by hand. 📍 The problem: Many property deeds in Northern California — especially in Santa Clara County — still contain illegal racial covenants from as far back as the 1850s, banning people of color from owning or living in certain homes. 🔎 The solution: The team used OCR to digitize 5.2 million deed pages, then fine-tuned Mistral-7B via LoRA on a hand-labeled dataset of discriminatory phrases. The model learned to identify harmful language in context, avoiding false positives like the surnames "Black" or "White." 📊 Results: Detected 24,500 lots with racial clauses (~25% of homes in 1950) Found that 10 developers were responsible for 1/3 of all discriminatory clauses Model reviewed all pages in 6 days at a cost of $258 Equivalent manual effort: 10 years, $1.4M 💥 This work doesn't just use LLMs to build the future — it uses them to understand and repair the past. The open-sourced model now enables other U.S. counties to follow suit. #AI #MachineLearning #LLM #ResponsibleAI #Stanford #Princeton #HistoryMeetsAI #TechForGood #EthicalAI #SocialImpact #PropertyTech https://lnkd.in/dtec-zbx

4