Ranjan Sinha, Ph.D.

New from Packt Publishing >> "Building Neo4j-Powered Applications with LLMs: Create LLM-driven search and recommendations applications with Haystack, LangChain4j, and Spring AI" by Ravindranatha Anthapu and Siddhant Agarwal Available at https://amzn.to/4l9lLKO This LLM book is for database developers and data scientists who want to leverage knowledge graphs with Neo4j and its vector search capabilities to build intelligent search and recommendation systems. Working knowledge of Python and Java is essential to follow along. Familiarity with Neo4j, the Cypher query language, and fundamental concepts of databases will come in handy. 𝓦𝓱𝓪𝓽 𝓨𝓸𝓾 𝓦𝓲𝓵𝓵 𝓛𝓮𝓪𝓻𝓷: 🟠Design, populate, and integrate a Neo4j knowledge graph with RAG 🔵Model data for knowledge graphs 🟢Integrate AI-powered search to enhance knowledge exploration 🟠Maintain and monitor your AI search application with Haystack 🔵Use LangChain4j and Spring AI for recommendations and personalization 🟢Seamlessly deploy your applications to Google Cloud Platform

42

3 Comments

Over the holiday break, I spent some time revisiting some of the foundational work shaping the field of scientific AI. One paper that stands out—published back in 2021—was co-authored by Damien Guého, now a Senior R&D Engineer at Geminus. The paper tackles a timeless challenge: how do you uncover the true physics of a system when all you have is sparse, noisy, imperfect data? Damien and his co-author's approach is both clever and elegant. Instead of a brute force deep learning strategy, they leveraged clever math to uncover a physics-grounded structure with interpretable dynamics—from messy, nonlinear signals. The result is a robust, reliable and generalizable solution that uses a fraction of the computational effort. They first solved the problem using a deep learning approach that required 7,720 parameters. They then applied their method that used just 3 terms and was 10,000× more accurate. This mindset is core to Geminus' mission to simplify engineering and operational decision making in complex systems where uncertainty abounds and data is sparse and messy. Work like that of Damien and his colleagues continues to inspire where we’re headed next. Proud of the team at Geminus Labs—and excited for what’s coming. Check out his paper here. https://lnkd.in/eHt-5tXX Watch this space - be sure to follow Geminus to be part of the conversation.

32

DeepSeek AI published a technical paper a few days ago, co-authored by co-founder Liang Wenfeng that argues for a new architecture to train foundational models. Essentially moving away from 'muscle head logic'. As the writer of the article I linked below by Larry Dignan states, The DeepSeek paper is a bit wonky. But remember what DeepSeek did to the market a year ago when they released their LLM - it was very performant and low cost. The AI market to date has been driven by muscle head logic. The answer for everything in the last two years has been more muscle. In AI's case that approach means more GPUs, more farm land going away, more energy usage, more water and by all means more data centers, debt and capital expenditures. This approach and consumption is likely to change with deep technical architectural change including at the chip level. Price and performance matter. LLM giants are diversifying their chip bases. Google's TPUs and Amazon Web Services' Trainium are going to be threats to Nvidia as inference becomes the main event. It will be really interesting to see how the industry shapes up here - lots to follow and very interesting! News Article here with link to DeepSeek paper: https://lnkd.in/eN38mSAV

6

3 Comments

Another breakthrough in #AI was published earlier today by Sapient Intelligence, and the linked headline by VentureBeat says it all. https://lnkd.in/giYJtEb9 TL;DR, the new #HierarchicalReasoningModel (#HRM) reasons more like a human by quickly thinking about the best way to solve a problem and then thinking more slowly to solve the problem. This new method can be trained on an order of magnitude fewer training examples and can solve multi-step puzzles that simply befuddle #CoT #LLM s, like #ChatGPT and #Claude, ultimately requiring far fewer resources.

6

2 Comments

An interesting study on scaling models. Everyone knows the Scaling Law—the more parameters, the smarter the model. That law has driven the industry in recent years: GPT-3 at 175B, PaLM at 540B, Llama-3.1 at 405B. The logic is simple – if you want better, make it bigger. But Chinese researchers from Tsinghua asked a different question: what if we look not at absolute size, but at efficiency? How much “intelligence” is packed into each billion parameters? They introduced the concept of capability density – the model’s ability density. The idea is interesting: take a reference model, plot a “parameters → quality” curve, and then for any other model, compute how many parameters the reference would need to achieve the same result. The ratio of this “effective size” to the actual size is the density. Analyzing 51 open models from the last two years, they found a pattern they call the Densing Law: the maximum LLM density doubles roughly every 3.5 months. The numbers are striking. Llama-1 in February 2023 had a density of about 0.1. Modern models like Gemma-2-9B and MiniCPM-3-4B reach densities around 2.0. Over two years, efficiency has increased 20-fold. Practically, this means model size required to reach the same quality is falling exponentially. MiniCPM with 2.4B parameters, released in February 2024, performs on par with Mistral-7B from September 2023. In four months, the required size shrank threefold. Even more interesting are inference prices. GPT-3.5 in December 2022 cost $20 per million tokens. Gemini-1.5-Flash in August 2024—$0.075. A 266× drop in under two years. Moreover, cost is falling faster than density is rising—because inference infrastructure is improving in parallel. The authors make an intriguing observation about combining laws. The Densing Law captures algorithmic efficiency growth; Moore’s Law captures hardware power growth. Together, they yield a doubling of the “effective size” of the model you can run on a fixed device roughly every 88 days. This opens the path to full-fledged LLMs on smartphones and laptops! It’s also interesting that compressing models via pruning and distillation doesn’t always increase density. Compressed models often end up less efficient than natively small models – apparently due to insufficient post-training after compression. The authors’ main conclusion is fairly provocative: the parameter-count arms race is a dead-end strategy. Any SOTA model will be outclassed in price/performance by newer compact models within months. DeepSeek-V3 has already shown this path – instead of piling on parameters, they focused on an efficient architecture and achieved top-tier quality at lower cost. In essence, the Densing Law is a Moore’s Law analogue for LLMs. Moore described progress in hardware; here we’re talking algorithms and data. And as long as this law holds, every 3.5 months we get models twice as efficient as the previous generation. https://lnkd.in/d5qHtUFZ

2

Hierarchical Abstraction is emerging as a key technique for building more robust and efficient LLMs. Instead of processing information as a continuous stream, this approach models data at multiple levels – think individual details *and* summarized concepts. This allows AI systems to retain crucial context over longer interactions, improving reasoning and reducing errors in complex tasks. For enterprises, this translates to more reliable AI copilots, better decision-making in dynamic environments, and scalable automation of knowledge-intensive operations. 🚀 Ultimately, it’s about building AI that ‘understands’ rather than just ‘remembers’. 💡📈

4

2026: The Year Enterprise AI Gets a "Nerve Center" ➡️ The debate between LLMs and structured data is officially over. As we move into 2026, the real winners aren't just "using AI"—they are orchestrating it. I just finished an insightful piece by Bill Tanner featuring leaders from Graphwise and Propel, Inc , and the roadmap for the next 12 months is clear: Context is King, but Infrastructure is the Kingdom. Here are my 3 biggest takeaways on how the enterprise landscape is shifting: 1. Hybrid AI is the New Gold Standard We are moving away from the "neural intuition" of models alone. The future is Hybrid. By blending the creative power of LLMs with the precision of Knowledge Graphs, enterprises are finally solving the "hallucination" problem. In 2026, a model that can’t explain why it made a decision is a liability, not an asset. 2. The Rise of "Agentic" Ecosystems The era of standalone software is fading. We’re seeing a shift from static bots to multi-agent ecosystems that reason and collaborate. But here’s the kicker: these agents are only as good as their "shared memory." Knowledge graphs are becoming the digital nerve center that allows agents to act with traceable, auditable logic. 3. B2A: Business-to-Algorithm 📈 This is a fascinatng shift. Commerce is being redefined as AI takes over vendor discovery and selection. If your product data isn’t "algorithm-ready" (clean, structured, and semantic), your business might effectively become invisible to the digital buyers of 2026. Trust isn't just a feature anymore; it’s infrastructure. Companies that treat data governance as a strategic asset will turn their service departments from cost centers into high-margin revenue engines. What’s your take? Are you prioritizing agentic workflows this year, or is your focus still on foundational data cleanup? Let’s discuss in the comments. 👇 #EnterpriseAI #AgenticSoftware #KnowledgeGraphs #SaaS #DigitalTransformation #FutureOfWork #TrendsShaping2026 https://lnkd.in/gw6JNXfQ

25

1 Comment

The concept of neuromodulating agentic memory is helpful for defining "steering behaviours" in generative AI pipelines. Vince Kellen PhD proposed a #neuromodulation approach to agentic AI memory. This idea gives many opportunities to bundle explainable memory operations in rather complex spatial memory architectures. In this post, we discuss an emerging neuromodulation formulation created by recursive generative agents that reason about the memory space using neuroscience knowledge present in the context. The resulting "steering bundles" help add more explainable features to the semantic processes of memory usage. This is an important aspect if we need a human-in-the-loop workflow. We separate memory usage algorithms from vector space methods, which, while powerful, offer only a limited "flavor" of retrieval for agentic reasoning. #Aiagens #agenticmemory #generativeAI #AI #system2 https://lnkd.in/gcbGSr7t

8

Cost Analytics in NVIDIA’s AI Infrastructure At NVIDIA’s AI scale, every GPU hour has cost implications. Infrastructure analytics must answer: ◦ Which workloads underutilize compute? ◦ Where do cost spikes emerge? ◦ How does usage differ across customer segments? ◦ What trends signal inefficiency early? Cost observability turns infrastructure from an operational metric into a strategic one. Tracking usage patterns, anomaly signals, and segment-level consumption enables better allocation and transparency. Small efficiency improvements at scale can translate into significant financial impact. Open to discussions around analytics that improve infrastructure efficiency and cost visibility at NVIDIA.

11

3 Comments