About
I am a translator. I speak Geek, Business, and Human. I am passionate about all things…
Articles by Pramath
Activity
11K followers
-
Pramath Malik reposted thisPramath Malik reposted thisInside the Samsung Galaxy S25 Ultra. How a Smartphone Reveals the Hidden Architecture of the Global Chip Industry. Inside Samsung’s Galaxy S25 Ultra lives an entire ecosystem of semiconductors, each built for a specific role, each optimized under different constraints, and each produced by a different part of the global supply chain. At the center sits advanced-node silicon handling the heavy compute load; AI acceleration, graphics, and system orchestration. Surrounding it is a wide collection of mature-node chips responsible for power management, connectivity, sensing, audio, storage control, and RF performance. Samsung is one of the few companies in the world with deep in-house capabilities across design, memory, fabrication, and system integration. And yet, even here, the smartphone is not built by a single company. A teardown reveals a familiar list of contributors across the industry; logic designers, memory suppliers, RF specialists, sensor manufacturers, and analog experts, each operating in their own technological domain. This reinforces a fundamental truth of the semiconductor industry ✳️ A smartphone is not made by one company. It is assembled by a global network of specialists. ✳️ There is a long chain between a wafer moving through a cleanroom and a smartphone powering on for the first time. That chain spans continents, companies, and specialisations that must align with near-perfect precision. A teardown like this makes something clear. ✴️ Innovation in semiconductors is not isolated but deeply coordinated. ✴️ ❇️ And every finished device is the visible tip of a very deep, very global system. ❇️ Credit to Gidion V. Simbo and the Behind The Chip substack for the visual 👉 https://lnkd.in/g7HMUGs4 I share semiconductor, AI and technology insights every day. Follow me 👉 Andrew Chan Yik Hong. For strategic perspectives on geopolitics, global supply chains, industrial policy and the business of technology shaping the semiconductor ecosystem. Ring the bell 🔔 to stay connected to the latest shifts across the global ecosystem. 💬 If this post resonates with you, re-post, leave a comment or drop a like. I look forward to hearing your thoughts.
-
Pramath Malik shared thisPrashasti Singh | Divine Feminine | Comedy Special | HindiPrashasti Singh | Divine Feminine | Comedy Special | Hindi
-
Pramath Malik reposted thisPramath Malik reposted thisHow my codebase written entirely with Claude Code runs 😁
-
Pramath Malik reposted thisPramath Malik reposted thisThe Bloomberg piece on the narrowing US-China AI gap hits on a reality engineering leaders have been navigating in real time: the AI race is no longer just about peak capability—it’s about inference economics, distribution, and architectural pragmatism. Having spent the last decade-plus building and scaling large-scale infrastructure, here are a few technical and strategic observations on where the industry is heading: 1. The "Performance-per-Dollar" Metric Has Overtaken Peak Benchmarks For a long time, the Silicon Valley playbook was simple: throw more compute at training, scale parameters, and pass the API costs downstream. But when open-weight models deliver 80–90% of frontier reasoning at 10% of the cost, the engineering trade-offs shift dramatically. For complex agentic workflows—where a single task might trigger dozens of autonomous debugging and execution loops—high token costs act as an immediate scaling bottleneck. Teams aren't just optimizing for accuracy anymore; they're optimizing for unit economics. 2. Open-Weight Models Are Winning the Distribution Moat By making models downloadable and run-anywhere, Chinese labs are capturing massive developer mindshare. Being able to self-host, fine-tune locally, and deploy on sovereign infrastructure solves massive enterprise pain points around data control, lock-in, and unpredictable API pricing. Billions of open-weight downloads represent more than just usage—they represent an ecosystem lock-in that accelerates downstream innovation faster than gated APIs can keep up. 3. The Rise of Hybrid & Multi-Tiered Inference Routing We are rapidly moving away from single-model tech stacks. The emerging pattern for resilient systems architecture is multi-tiered model routing: Tier 1 (Frontier Models): Reserved for high-stakes, real-time user interactions, complex reasoning, or sensitive edge cases requiring strict guardrails. Tier 2 (Open-Weight / Budget Models): Heavy lifting, asynchronous batching, background agentic iterations, and internal tooling. Architecting intelligent gateway layers that dynamically route requests based on task complexity, latency budgets, and cost thresholds is fast becoming one of the most vital responsibilities for modern AI platform engineering. 4. Efficiency Spurred by Constraint When hardware constraints exist, algorithmic and systems engineering step up. The rapid improvement in quantization, distillation, and post-training optimization shows that raw FLOPS aren't the only leverage point. Breakthroughs in software efficiency can significantly level the playing field against sheer hardware scale. The Bigger Picture: We are entering an era of multi-polar AI development. A single lab maintaining an unquestioned lead is increasingly unlikely. The durable advantage won't just belong to whoever trains the largest model—it will belong to the platforms that offer the best deployment efficiency, developer velocity, and real-world system reliability.
-
Pramath Malik reposted thisPramath Malik reposted thisGoogle DeepMind undergoes major leadership shake up as Demis Hassabis steps back and Jeff Dean departs to start a new AI venture. - Hassabis moves from CEO of Google DeepMind to Chair + Chief Scientist of Alphabet, staying on as head of Isomorphic Labs - Koray Kavukcuoglu (formerly CTO) becomes SVP of Google DeepMind, now overseeing Gemini development and frontier AI research - Jeff Dean exits after 27 years to launch Discovery Loop, an independent venture focused on automating scientific research: joined by Sanjay Ghemawat, Oriol Vinyals, and Quoc Le Google says the two moves, though announced together, are unrelated. Alphabet shares dipped on the news. A significant shift atop one of the world's leading AI labs. Best, Ilir Aliu ------ Weekly robotics and AI insights. Subscribe free: 22astronauts.com
-
Pramath Malik reposted thisPramath Malik reposted thisJeremy Strong knows exactly what he’s doing. With his infamous brand of artistic commitment, our September cover star made us root for Succession’s biggest failson—and earned a place in prestige TV history. Now, with a key role in Aaron Sorkin’s upcoming The Social Reckoning, he’s taking on an even bigger challenge: Can he make us feel for Mark Zuckerberg? See the cover story here: https://lnkd.in/gS-W4ryA
Pramath Malik commented on a post
1w
Guardrails k andar hai na.. chill
-
Pramath Malik liked thisPramath Malik liked thisInside the Samsung Galaxy S25 Ultra. How a Smartphone Reveals the Hidden Architecture of the Global Chip Industry. Inside Samsung’s Galaxy S25 Ultra lives an entire ecosystem of semiconductors, each built for a specific role, each optimized under different constraints, and each produced by a different part of the global supply chain. At the center sits advanced-node silicon handling the heavy compute load; AI acceleration, graphics, and system orchestration. Surrounding it is a wide collection of mature-node chips responsible for power management, connectivity, sensing, audio, storage control, and RF performance. Samsung is one of the few companies in the world with deep in-house capabilities across design, memory, fabrication, and system integration. And yet, even here, the smartphone is not built by a single company. A teardown reveals a familiar list of contributors across the industry; logic designers, memory suppliers, RF specialists, sensor manufacturers, and analog experts, each operating in their own technological domain. This reinforces a fundamental truth of the semiconductor industry ✳️ A smartphone is not made by one company. It is assembled by a global network of specialists. ✳️ There is a long chain between a wafer moving through a cleanroom and a smartphone powering on for the first time. That chain spans continents, companies, and specialisations that must align with near-perfect precision. A teardown like this makes something clear. ✴️ Innovation in semiconductors is not isolated but deeply coordinated. ✴️ ❇️ And every finished device is the visible tip of a very deep, very global system. ❇️ Credit to Gidion V. Simbo and the Behind The Chip substack for the visual 👉 https://lnkd.in/g7HMUGs4 I share semiconductor, AI and technology insights every day. Follow me 👉 Andrew Chan Yik Hong. For strategic perspectives on geopolitics, global supply chains, industrial policy and the business of technology shaping the semiconductor ecosystem. Ring the bell 🔔 to stay connected to the latest shifts across the global ecosystem. 💬 If this post resonates with you, re-post, leave a comment or drop a like. I look forward to hearing your thoughts.
-
Pramath Malik liked thisPramath Malik liked thisHow my codebase written entirely with Claude Code runs 😁
-
Pramath Malik liked thisPramath Malik liked thisTalent is flowing. TL;DR: Many of the people who built the last generation of generational companies are now building the current generation of generational companies. The similarities are interesting markers of ongoing talent wars, primarily for top technical talent. The differences are interesting as potential signs of where each company is focused. For the long-time followers, you'll know a familiar line from my time at Live Data Technologies (data source): "Every hire is a $XXXk/yr bet on what a company is building and selling."
-
Pramath Malik liked thisPramath Malik liked thisThe Bloomberg piece on the narrowing US-China AI gap hits on a reality engineering leaders have been navigating in real time: the AI race is no longer just about peak capability—it’s about inference economics, distribution, and architectural pragmatism. Having spent the last decade-plus building and scaling large-scale infrastructure, here are a few technical and strategic observations on where the industry is heading: 1. The "Performance-per-Dollar" Metric Has Overtaken Peak Benchmarks For a long time, the Silicon Valley playbook was simple: throw more compute at training, scale parameters, and pass the API costs downstream. But when open-weight models deliver 80–90% of frontier reasoning at 10% of the cost, the engineering trade-offs shift dramatically. For complex agentic workflows—where a single task might trigger dozens of autonomous debugging and execution loops—high token costs act as an immediate scaling bottleneck. Teams aren't just optimizing for accuracy anymore; they're optimizing for unit economics. 2. Open-Weight Models Are Winning the Distribution Moat By making models downloadable and run-anywhere, Chinese labs are capturing massive developer mindshare. Being able to self-host, fine-tune locally, and deploy on sovereign infrastructure solves massive enterprise pain points around data control, lock-in, and unpredictable API pricing. Billions of open-weight downloads represent more than just usage—they represent an ecosystem lock-in that accelerates downstream innovation faster than gated APIs can keep up. 3. The Rise of Hybrid & Multi-Tiered Inference Routing We are rapidly moving away from single-model tech stacks. The emerging pattern for resilient systems architecture is multi-tiered model routing: Tier 1 (Frontier Models): Reserved for high-stakes, real-time user interactions, complex reasoning, or sensitive edge cases requiring strict guardrails. Tier 2 (Open-Weight / Budget Models): Heavy lifting, asynchronous batching, background agentic iterations, and internal tooling. Architecting intelligent gateway layers that dynamically route requests based on task complexity, latency budgets, and cost thresholds is fast becoming one of the most vital responsibilities for modern AI platform engineering. 4. Efficiency Spurred by Constraint When hardware constraints exist, algorithmic and systems engineering step up. The rapid improvement in quantization, distillation, and post-training optimization shows that raw FLOPS aren't the only leverage point. Breakthroughs in software efficiency can significantly level the playing field against sheer hardware scale. The Bigger Picture: We are entering an era of multi-polar AI development. A single lab maintaining an unquestioned lead is increasingly unlikely. The durable advantage won't just belong to whoever trains the largest model—it will belong to the platforms that offer the best deployment efficiency, developer velocity, and real-world system reliability.
Experience & Education
Courses
-
Application of Game Theory to Business
AGTB
-
Management of Information Systems
MIS
-
Marketing Management I & II
MMI & II
-
Product Management
PM
-
Sales & Distribution Management
SDM
Projects
-
Eyes On Demand
During Startup Weekend MEGA Robotics division our team worked on developing a device useable by blind or visually impaired people. It will provide a set of augmented reality tools to help the blind in their daily life. (HW+SW)
We were awarded first prize amongst 100s of competitors
https://www.youtube.com/watch?v=7YMYaQufHP8Other creatorsSee project
Languages
-
English
Native or bilingual proficiency
-
Hindi
Native or bilingual proficiency
Recommendations received
9 people have recommended Pramath
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content