Ali A.

✨ Imagine this: You’re talking to an AI, and instead of forgetting everything after the chat ends, it remembers. It recalls past conversations, learns from them, and builds context like a trusted partner. That’s precisely the world I stepped into with the LLMs as Operating Systems: Agent Memory course from DeepLearning.AI. . 🚀 At first, it felt almost sci-fi: The LLM’s context window became its short-term memory. External storage turned into long-term memory. And the agent itself? It was like an operating system, intelligently deciding what to keep, what to forget, and when to retrieve information. 🧠 The highlight for me was experimenting with MemGPT-style agents. These aren’t just chatbots — they can self-edit their memory, update knowledge, and even collaborate with other agents using shared memory. Suddenly, “AI memory” wasn’t just storage — it was strategy. And the “aha!” moment? When I saw how selective memory works. It’s not just about remembering — it’s about remembering responsibly like an AI that recalls patterns but knows when to redact personal details depending on security or authority. That shift made me realize that memory is the key to trusting AI. This course didn’t just teach me a framework — it gave me a glimpse into the future of AI agents as true operating systems. Agents that don’t just answer, but grow with you. I walked away thinking: 👉 The difference between a chatbot and a real agent isn’t just intelligence — it’s memory. And now, I can’t stop wondering — how will memory-driven agents reshape industries like healthcare, education, or even project management? What do you think — are we ready for AI that truly remembers? #AI #AgenticAI #LLMs #ArtificialIntelligence #MemGPT #FutureOfWork

25

1 Comment

Congratulations Pia Saunders Campbell, Merve Ayyüce KIZRAK, Ph.D., and Elsie Jang on the publication of "Meeting the AI Workforce Challenge." This blend of public-private-philanthropic perspectives, is a great read for anyone engaged in workforce development, AI policy, or the future of talent. As with any of the prior technological breakthroughs, we must and will evolve into the new jobs and skills of the future; however, the rapid pace of AI calls on us to do this faster than ever before. Key Takeaways 👇 🔄 AI is rapidly reshaping the U.S. labor market, particularly in entry-level and technical roles. Adapting quickly is urgent for both organizations and individuals. 🤝 The federal government's AI Action Plan and America’s Talent Strategy are solid starts, but coordinated cross-sector action—including youth leadership and local partnerships—is essential for equitable, future-ready outcomes. 📚 There’s a pressing need to integrate AI literacy, ethical reasoning, and problem-solving into all levels of education and professional training, supporting resilient, adaptive workers. 🪴 Youth engagement and inclusion are critical. Policies shaped with (and by) young people will be more innovative and enduring as the pace of technological change accelerates. Grateful for the opportunity to share my perspectives with this team along the way and learn alongside such passionate change makers! https://lnkd.in/eKfNX7UC #AI #WorkforceDevelopment #FutureofWork #Collaboration #TalentStrategy

21

“The only scalable way to measure complex metrics like hallucination, conciseness is by using another LLM tuned and trained to evaluate your model’s responses.” That was Oguzhan (Ouz) Gencoglu Gencoglu, Co-Founder & Head of AI at Root Signals, on this week’s episode of the (AI) People podcast. We explored why traditional MLOps metrics can’t capture what really matters in LLM-powered products. Ouz believes LLM-as-a-Judge is not just useful it’s the only way to get reliable, scalable evaluation without relying on armies of humans. 👉 Would you trust an LLM to judge another LLM’s output?

16

7 Comments

The course on AI Agents (taught at UCSD CSE) is comprehensive, covering everything from classical planning and reinforcement learning to cutting-edge topics like multi-agent systems and human-AI collaboration. I could only find slides and suggested reading (no video, but perhaps someone uploaded it) Course Link: https://lnkd.in/gSknkQKG

71

Stanford’s CME 295 is one of the most practical LLM courses available today. These 8 lectures explain how modern LLMs work, from core concepts to agents and evaluation. If you want depth without noise, this series is worth your time. Lecture 1: End-to-End LLM Overview - Covers the full LLM stack, from transformers and training to agents, vision models, and evaluation. 🔗 https://lnkd.in/dGnQW39t Lecture 2: Transformer Internals and Optimizations - Explains positional embeddings, attention improvements, normalization, and BERT architecture in depth. 🔗 https://lnkd.in/dT_VEpVH Lecture 3: Large Language Models in Practice - Focuses on LLM design, MoE models, decoding methods, prompting, and inference optimizations. 🔗 https://lnkd.in/dwjjpjaP Lecture 4: LLM Training and Finetuning - Breaks down pretraining, scaling laws, optimization techniques, and efficient finetuning methods. 🔗 https://lnkd.in/dSi_xCEN Lecture 5: Preference and Alignment Tuning - Walks through RLHF, reward models, PPO, and modern alternatives like DPO. 🔗 https://lnkd.in/dUK5djpB Lecture 6: LLM Reasoning - Covers reasoning benchmarks, reinforcement learning scaling, and advanced optimization methods. 🔗 https://lnkd.in/dAGQTNAM Lecture 7: Agentic LLMs and RAG - Explains retrieval systems, tool calling, agent frameworks, and safety considerations. 🔗 https://lnkd.in/dWD4j7vm Lecture 8: LLM and Agent Evaluation - Details evaluation metrics, LLM-as-judge methods, and benchmarks for reasoning, coding, and safety. 🔗 https://lnkd.in/ddxE5zvb Start with this today. Follow Shubham Vora to learn more about developing AI agents and how to implement them in your business.

84

44 Comments

🧪 A key challenge to solving #InverseProblems, especially in science and engineering settings, is limited data and compute for simulations. 🥧 The #ICLR2025 work of Apivich Hemachandra, Gregory Lau, See-Kiong Ng, and Bryan Kian Hsiang Low, PIED, is an #ExperimentalDesign framework for #PDE inverse problems that efficiently optimizes for the best observation inputs. 📌 PIED utilizes #PINNs to perform forward simulation and to solve the inverse problems, which allows for higher efficiency over numerical simulators and allow for meshless and differentiable simulations. 🤖 Using PINNs along with differentiable observation selection criteria, PIED selects the optimal design parameters for one-shot deployment, while exploitation parallel computation and gradient-based optimization methods. 📈 These benefits allow PIED to outperform existing ED benchmarks in IPs for both finite-dimensional and function-values inverse parameters, as demonstrated in various settings including on real experimental data. Check out our #ICLR2025 Poster #28 on 26 Apr in Poster Session 6 Sat 3pm Hall 3 + Hall 2B. Paper: https://lnkd.in/gDxav2Dn GitHub: https://lnkd.in/gZXDkA5H

31

1 Comment

MISRAJ AI introduces Baseer (بصير), a specialized vision-language model that sets a new state-of-the-art for the notoriously difficult task of Arabic document Optical Character Recognition (OCR). By meticulously fine-tuning a general-purpose Multimodal Large Language Model (MLLM) on a massive, curated dataset, the researchers have created a model that not only extracts text but also understands document structure, significantly outperforming existing open-source and large-scale commercial systems. The creation of Baseer was a multi-stage process centered on data and an efficient training strategy: 1. Massive Hybrid Dataset Curation: The team built a 500,000-pair dataset specifically for Arabic document-to-markdown OCR. This dataset is a hybrid of: ✅ 300,000 Synthetic Documents: Created using an in-house pipeline to generate a vast diversity of layouts, fonts, and document types, including aggressive data augmentation to simulate real-world degradation. ✅ 200,000 Real-World Documents: Curated from a diverse collection of books, magazines, and academic papers to capture authentic layout complexities and typographic conventions. 2. Efficient Decoder-Only Fine-Tuning: The team chose the Qwen2.5-VL-3B-Instruct model as their foundation due to its strong baseline performance in Arabic. Crucially, they adopted a decoder-only fine-tuning strategy, where the pre-trained vision encoder was kept frozen. This preserved the model’s powerful, generalized visual features while allowing the language decoder to specialize entirely on the nuances of Arabic text and document structure. 3. A New, High-Quality Benchmark: To rigorously evaluate their work, the researchers also created and open-sourced Misraj-DocOCR, a new expert-verified benchmark for Arabic OCR, addressing significant errors and a lack of diversity in existing benchmarks. This is the key diagram from the paper, providing a clear and concise overview of the entire data and training pipeline, from the hybrid dataset to the final fine-tuned Baseer model. ➡  You can read the paper from here: https://lnkd.in/dhjaceCg ------------------------------ ➡ Do not miss the 50% discount on the To Data & Beyond yearly subscription: https://lnkd.in/dZ2Sjkhk ➡ All My 7 Books, One Button Away, With 40% Off: https://lnkd.in/dqzXcDyJ #llm #opensource #largeLanguagemodels

50

2 Comments

Seemingly Conscious AI: Don’t Confuse the Tool for a Person Mustafa Suleyman warns that within a few years, we may face AI that seems conscious—not truly sentient, but convincing enough to mislead people. That illusion is dangerous. It risks creating emotional dependence, blurring moral lines, and even fueling calls for “AI rights.” My concern: when AI is allowed to impersonate a person, we weaken human dignity and confuse society’s foundations. AI should serve people, not pose as people. We need clear guardrails now—before habits and norms are set—to keep AI useful, not uncanny. Let’s preserve clarity: AI is a tool, not a soul. 👉 What do you think—should AI ever be designed to appear “conscious,” or is that a line we shouldn’t cross? #ArtificialIntelligence #EthicsInAI #FutureOfWork #TechnologyLeadership #ResponsibleAI https://lnkd.in/eycpNMnC

18

2 Comments

Architectural breakthroughs in Gen AI continue! Moonshot AI’s Kimi K2 has achieved top leaderboard positions, surpassing GPT-4, thanks to key innovations—especially the introduction of the MuonClip optimizer. This new approach stabilizes training for massive Mixture-of-Experts (MoE) models, significantly boosting performance and efficiency at unprecedented scale. https://lnkd.in/gvYrGzTK

24

### What Cross-Entropy Teaches LLMs ### Cross-entropy loss is so foundational in language modeling that we rarely pause to question it. It is usually presented as a neutral objective. Penalize incorrect predictions, reward correct ones, and intelligence will emerge. But this view is incomplete. As I explore in my recent blog (https://lnkd.in/gazzvwap), What Cross-Entropy Teaches Large Language Models. Cross-entropy does more than measure error. It applies behavioral pressure. It shapes how models resolve uncertainty, not just whether they predict correctly. Over time, this quietly teaches models to commit early, even in situations where hesitation would be the more intelligent response. We optimize for confidence and then act surprised when hallucinations appear. Perhaps the deeper question is not only how accurate our models are, but how we train them to live with uncertainty. #ArtificialIntelligence #LargeLanguageModels #MachineLearning #ModelBehavior #Uncertainty #AIResearch Note: These blogs aim to be intellectually stimulating, focusing on the science and intuition behind algorithms. No step-by-step walkthroughs. Just thoughtful, quick, and fun reads.

22

🎥 Watch: Why are Arabic LLMs still navigating a "data desert"? 🏜️ Back in 2022—well before the current hype—my team identified the exact technological barriers and data scarcities that keep Arabic AI in a "catchup" phase compared to English-centric models. In this short video, "The Race for Arabic AI," I explain why English LLMs dominate and how our foundational work with NLP expert Wajdi Zaghouani Ph.D put us ahead of the game. 📖 Our 2022 Research: https://lnkd.in/emJhX7Uy https://lnkd.in/dg2dCzsF authors: Nashva Ali Wajdi Zaghouani Ph.D Mahmood Alzubaidi (PhD) Prof. Mowafa Househ Alaa Abd-alrazaq, PhD, FIAHSI, FACHDM

22

1 Comment

Get faster and smarter MOE inference straight out of the box. 👇Deep dive on scaling expert parallelism with TensorRT-LLM. LLMs with MOE promise higher model capacity without linearly increasing compute costs - but they introduce new challenges -- more conditional computation, dynamic routing, and non-uniform GPU utilization -- solved by TensorRT-LLM. ✨ New in TensorRT-LLM has native support for expert parallelism—designed for fast, efficient inference with MoE models like Mixtral (Mistral AI) and DeepSeek (DeepSeek AI). This gives you: ✅ Dynamic expert routing: Automatically route tokens to the top-k experts with minimal overhead. ✅ Efficient expert scheduling: Balance expert loads across GPUs using smart sharding and token bucketization. ✅ Memory-aware execution: Maximize hardware utilization while respecting memory budgets. ✅ Drop-in support: Use @HuggingFace models with minimal code changes via TensorRT-LLM's #Python API. 🧠 How it works: MoE models activate only a subset of "experts" for each token. This dynamic nature is powerful—but hard to optimize. It’s all done under the hood using custom #CUDA kernels and NCCL-based communication primitives—giving you low latency, high throughput, and better GPU scaling. ✨ TensorRT-LLM handles: ✅Token-expert mapping using the gating network. ✅Token sorting to batch same-expert tokens together. ✅Expert parallel execution across GPUs. ✅Merging outputs for final predictions. 🛠️ Developer workflow - here is the code to get started. # Clone the repo git clone https://lnkd.in/g-GiDX23 # Use included examples to load and run a Mixtral model cd TensorRT-LLM/examples/mixtral From there, the Python API lets you load the model, convert it with TensorRT, and run expert parallel inference—all with a few lines of code. Results? 📈 Performance at scale. Tests show up to 2.3x faster inference throughput compared to standard tensor parallelism when using 8 GPUs and top-2 experts per token. Even better—TensorRT-LLM keeps efficiency high across increasing batch sizes. Want to see it in action or contribute? 👉 Read the full tech blog: https://lnkd.in/g_7YV3vV 👉 Explore the code on GitHub: https://lnkd.in/gNjQ5W2U 👉 Follow updates in the TensorRT-LLM repo: https://lnkd.in/gqSHYQ4u Share your experiences with us.

78

3 Comments

Doing more of what I love recently: timbre modelling of the Tonbak. The Tonbak (or Zarb) is an Iranian goblet drum, widely regarded as the principal percussion instrument of Persian music. I’ll spare you the real-time timbre transfer explanation, since you’re probably already familiar with the concept. If not, just google RAVE and DDSP, read a bit of the material, and try training something yourself. What interests me here is twofold. First, can we define certain input sound characteristics that would deliberately target a specific latent trajectory, ultimately producing a plausible and musically useful output? In other words, what if we could design, or even synthesize, input signals that, when triggered and use for inferring these models, consistently generate desirable sonic qualities? The other one is complements the former, how far could we take this algorithmically by employing complex pre/post processing DSPs. Anyways, I’ll leave you with this documentation snippet for your to hear some yourself. It feels nice to revisit this project. #computermusic #nueralsynthesis #aimusic

103

9 Comments