Gaurav Singh

Satya Nadella asked for "a new equilibrium in terms of our 'theory of the mind'" that accounts for humans equipped with AI. He wants to evolve "bicycles for the mind" into something precise enough to guide product design. Here's a candidate: Human knowing operates through four irreducible operations—experience, understanding, judgment, decision. AI amplifies the first, assists the second, and cannot perform the third or fourth. Or more simply: AI is a bicycle for experience, a workshop for understanding, and a null for judgment and decision. This isn't pessimism about capability. It's precision about kind. And it dissolves the slop-versus-sophistication debate entirely—both sides argue output quality when the real question is operational integrity. Nadella's post: https://lnkd.in/gPg9EJAM My response: https://lnkd.in/gueiDMMk

33

8 Comments

https://lnkd.in/gAhkuzQr While this sounds like an ambitious goal to have, cetainly achievable with the current momentum we are seeing in the country. Having involved with some of the state’s leadership in early discussions to support and enable deeptech ecosystem, this is one major shift destined to change India tech output worldwide. To qualify this at foundational level, Indian energy ecosystem is going through massive transformation not just with #energy #transition and #digitalmapping but also at generation level. We have been experiencing the shift towards #solar, #hydro, #wind and other efficient ways to address the shot up demand in coming days which in my opinion may not compensate for the rapid development we are aiming at. Very much expecting some #revolutionary #policy level decisions around #nuclearfission or similar high end play in coming days. In short, #IndiaAI revolution require a friendly #nuclearenergy #scaleup to sustain it. #ViksitBharat #Bharat2030 #Deeptech Ministry of Electronics and Information Technology

24

Though not reported, AWS's AI growth and traction seemingly trailed peers in 2026, partially due to capacity constraints, lack of leading-edge proprietary frontier models, and limitations on Trainium capabilities (plus added upfront costs per our checks). Capacity challenges to compete for AI workloads should be less of a factor in 2026, and we think the narrative on AWS's AI positioning could shift in 2026 given recent management changes that signal a commitment to AI innovation.

2

🚀 Google Just Unveiled a New Wave of AI Superpowers 🤖💡 Why this matters: Google is going all in on AI — not just for developers, but for everyone who creates, learns, or builds. From making apps to generating videos or chatting with YouTube itself — these new tools are shaping how the world works with AI. 🌍⚡ ⸻ Here are some of the most powerful new tools: 💻 Firebase Studio – Build and launch web or mobile apps faster with AI guidance. 🎬 Veo (Video Generation) – Turn text prompts into cinematic short videos or animations. 🎥 Gemini Ask on YouTube – Chat directly with videos for answers, insights, or summaries. 🧠 Gems in Gemini – Create your own custom AI assistants using your data and files. 🎨 Nano Banana (Editing) – Edit or enhance AI-generated images — or create new versions instantly. 📊 Gemini in Google Sheets – Let AI write, analyze, and generate formulas inside Sheets. ⚙️ Google App Builder – Build full apps from prompts — no code required. 🖼️ Media Generation (Imagen / Nano Banana) – Create stunning visuals from simple text. 🗣️ Gemini Live (Stream) – Chat with AI live, share your screen, and co-create in real time. 🧪 Google AI Studio – Test and compare Google’s models, prompts, and responses. 📚 NotebookLM – Transform notes into mind maps, summaries, quizzes, or even audio lessons. ⸻ Why it’s big: ✅ Brings creativity, data, and AI under one ecosystem ✅ Turns every user into a creator or builder ✅ Reduces barriers to code, design, and productivity ✅ Positions Google as the hub for next-gen AI tools ⸻ In short: Google just made AI creation mainstream. From classrooms to boardrooms — the Gemini era is here. 🌐🔥 #AI #Google #Gemini #Innovation #AITools #AIEducation #AICreation #GoogleLabs #AIRevolution #TechNews

8

Did you hear it? … the song of the AI Agents dominating #awsreinvent2025 Nearly every other booth you walk to is demoing an #aiagent platform. In a nutshell, we’re talking about simple ai agents. I’d call them baby agents. A prompt, a RAG, and a few MCPs here and there to collect data, all governed by an agent studio to design agent workflows. Most platforms are promoting an open model framework where you can pick and choose your preferred LLM. Bedrock is one of the key options available. What’s missing? The business application. A lot of those platforms are in their very early stages, many are not even GA yet. What’s common: build from scratch mentality, and my ai agent platform can do everything, everywhere, all at once. IT buyers? Torn down between all of these platforms not knowing if it’s the right time to start choosing one or many. The potential? It’s going to be insanely huge, with businesses and organizations reaching incredible levels of automation, while also providing superior quality and experience to their customers. Get ready as 2026 will be exciting! [sounds surreal to say so, nearly have to apologize that my post did not use any single AI, not even for brainstorming or improving my style. WDYT: Better with or without AI?]

16

4 Comments

There are two common RL training formulations for LLMs: Markov Decision Process (MDP) formulation and bandit formulation. Here’s how they work… Background: We should recall that an LLM generates output via next token prediction; i.e., by sequentially generating each output token. As shown in the image, the next token prediction process maps very easily to an RL setup—we can just model each token as an individual action! MDP formulation: The approach of modeling each token in the LLM’s output as an individual action is called the Markov Decision Process (MDP) formulation. An MDP is simply a probabilistic framework for modeling decision-making that includes states, actions, transition probabilities and rewards—this is exactly the setup used by LLMs for RL! Our initial state is the prompt and our policy acts by predicting tokens. During generation, actions are taken by selecting a token from the LLM’s predicted token distribution—each token is its own action. After a token is predicted, it is added to the current state and used by the LLM to predict the next token—this is just autoregressive next token prediction! Eventually, the LLM predicts a stop token (e.g., <|end_of_text|> or <eos>) to complete the generation process and yield a full trajectory. Outcome vs process reward: In the MDP formulation, we depict a reward being provided to our policy at every step. But, the reward mechanism for an LLM is usually a bit different. Most LLMs are trained using outcome supervision, meaning that a reward is only assigned after the model has generated a complete response (i.e., after the <eos>). Bandit formulation: In an outcome supervision setting, we may begin to question the utility of modeling each token as its own action. How do we handle credit assignment in this scenario? Alternatively, we could model the entire response as a single action that receives an outcome reward. This is the key idea behind the bandit formulation for RL training. This name comes from the idea of a contextual bandit in probability theory. The bandit setup is simple: our agent chooses an action, receives a reward and the episode ends. Our complete trajectory is a single action and reward! For LLMs, our action is the full completion generated for a prompt, which receives an outcome reward. Which formulation is better? In the context of LLMs, we can easily compute the probability of individual tokens or a full completion to a prompt, so we have the ability to model RL using either an MDP or bandit formulation. Given that LLMs usually only receive outcome rewards, however, the bandit formulation—despite being very simple—is quite fitting for LLMs. Plus, it tends to be more simple (and has been argued to perform similarly despite this simplicity). However, MDP formulations are also commonly used. For example, PPO uses an MDP formulation, whereas REINFORCE and RLOO adopt a bandit formulation. In other words, both RL formulations are viable and are used for training LLMs.

181

6 Comments

🚀 BIG NEWS in AI! Indian-origin company Uniphore just raised a massive $260 million (Series F) from global tech giants NVIDIA, AMD, Snowflake, and Databricks — pushing its total funding close to $870 million! 💰 Founded by Umesh Sachdev and Ravi Saraogi, Uniphore has become one of the world’s leading Enterprise AI platforms — seamlessly blending Conversational AI, Emotion AI, and Generative AI for businesses worldwide. 🌍 💡 Why this matters: 🔹 Enterprise AI is no longer experimental — it’s going mainstream. 🔹 Backed by chip, cloud, and data-analytics giants, Uniphore is poised to redefine how organizations automate customer and employee interactions. 🔹 A proud moment for Indian innovation making global impact! 🇮🇳 🔥 Key Investors: ✅ NVIDIA ✅ AMD ✅ Snowflake ✅ Databricks 💬 What this signals: The future of business AI will be agentic, composable, and deeply integrated into every workflow — and Uniphore is leading that transformation. 👏 Congratulations to the entire Uniphore team, Umesh Sachdev, and Ravi Saraogi for putting India on the global AI map once again! 👉 Read the full story here: https://lnkd.in/gbsWfg-E #Uniphore #AI #EnterpriseAI #StartupFunding #Innovation #NVIDIA #AMD #Snowflake #Databricks #TechNews #ScoopEarth #IndianStartups #UmeshSachdev #RaviSaraogi #scoopearthmagazine

12

LLMs display human-like behavior, with Karpathy once describing them as stochastic "people spirits." This makes them notoriously hard to benchmark, leading most teams to skip this step entirely and ship a poorly optimized model. We wrote a two-part series on performance benchmarking we wish existed when we started. In this first chapter, learn how to run InferenceMAX and build your own benchmark tailored to your workload. Read it here: https://lnkd.in/g34kpqDp

17