Puneet Kamal

What if your queries required access to multiple sub-related data from different documents, for an LLM to parse and make sense of before returning a coherent and accurate output? We often ask complicated questions, right? 🔹Well to solve this problem, these researchers released Multi-Head RAG (MRAG) which leverages activations of Transformer’s multi-head attention layer, instead of the decoder layer, as keys for fetching multi-aspect documents. They are taking advantage of the fact that multi-head attention can learn to capture different data aspects. 🔹The interesting part is that MRAG is both cost-effective and energy-efficient. In other words, it does not require additional LLM queries, multiple model instances, increased storage, or multiple inference passes over the embedding model. 🔹Moreover, MRAG can be seamlessly integrated with existing RAG frameworks and benchmarking tools like RAGAS as well as different classes of data stores. 🔹The researchers provide a comprehensive evaluation methodology, including specific metrics, synthetic datasets, and real world use cases to demonstrate MRAG’s effectiveness. 🔸Do check it out! #genai #technology #artificialintelligence Click #linkedangle and follow for more content Set the 🔔 notification on my page, don’t miss a post!

181

10 Comments

⭐ The Evolving Role of Product Management in the Age of AI With AI transforming the landscape, product management is undergoing a radical shift. Insights from recent discussion(link in comments) from at Lenny's summit with CPOs at Anthropic and OpenAI highlighted some key areas PMs need to master to build effective AI products. Here are five core insights I would like to share - 1. Mastering Evaluation - The effectiveness of an AI feature depends on how well its evaluations are crafted. - PMs must go beyond surface metrics and understand why and where models fail. Deep data analysis is now a critical skill. 2. Prototyping & Technical Know-How - Rapid prototyping with AI tools like Claude speeds up development by enabling quick A/B testing of different UI options. - A deeper technical understanding helps PMs better collaborate with research teams and make informed choices on features. 3. Designing for Non-Determinism - AI introduces unpredictability. Unlike traditional software, the same input doesn’t always yield the same output. - PMs must design adaptable products, incorporating feedback loops and guardrails to keep outcomes within desirable bounds. 4. The 'Model Whisperer' - The boundary between research PMs and product PMs is blurring, giving rise to a new breed of PMs with a model-centric approach. - Knowing how to create effective evaluations and prompts is increasingly crucial. 5. Preparing for the Future - The next generation of AI will be proactive, asynchronous, and multimodal. PMs should start planning for these experiences - where AI anticipates needs, works in the background, and interacts through text, voice, and visuals. AI is changing our world, and as product leaders, we must evolve with it. If you are a PM or a Product Leader, lets embrace these shifts to stay ahead. How are you preparing for the future - comment below 👇 or message me 💬 to partner in this journey 🚀 #ProductManagement #AI #FutureOfWork #Innovation #ProductXperts

20

2 Comments

BharatGPT by CoRover  OpenHathi by Sarvam.ai Dhenu by Kissan AI Krutrim by Ola These are a few examples of LLM models launched by Indian companies and startups recently, showcasing the rising trend of India-focused LLMs in the AI landscape. My question today is - How is Indian Data contributing to the development of these LLM Models? Today, Indian contextualised data is being used to develop LLM models that can understand and generate content in Indian languages, including Hindi, Tamil, Telugu, Bengali , Marathi etc.. These models are being used for a wide range of applications - including customer service, marketing, compliance, to develop personalised recommendations, language translations, and education tools that can cater to the unique needs and preferences of Indian consumers. Indian consumers produce a wide range of data that is being used for the development of LLM models. This data includes text, audio, and video content generated through social media, e-commerce, digital entertainment, and other online platforms. ⭕ India is the second-largest generator of digital data, after China. ⭕ The Indian AI market is projected to surpass $17 billion by 2030, growing at a CAGR of 48%, driven by the adoption of AI across various industries. ⭕ Today, LLM models are being trained on Indian data to facilitate data localisation, which is important for compliance with the Digital Personal Data Protection (DPDP) regulation in India and support GOI's National Strategy for AI. In addition, specifically setup Indian Data Exchanges play a crucial role in facilitating the sharing and access of data related to urban governance, service delivery, and smart city initiatives. 1. India Urban Data Exchange (IUDX): serves as a platform for data providers and users to share, request, and access datasets related to cities, urban governance, and urban service delivery. 2. Open Data Platform (ODP): It provides a single point of access to open datasets available with various smart cities, ministries/departments, and enhances visualization and discoverability of resources. To access this data, researchers and developers can leverage APIs provided by various online platforms, social media networks, and other digital services. It is important to ensure that data collection and use are complying with relevant laws and regulations, including global and local data privacy and protection laws. Today, India's large and diverse population, educational institutions, and IT companies are providing a rich source of data and talent that can be used for the training of LLMs and the development of Gen AI that can be applied to both global and local challenges. However, in my view, India still needs to focus on incentivising AI research and development, and increasing the number of AI and machine learning specialists to fully realise the potential of this technology. More on this in the coming days. Hope you found this useful.

20

2 Comments

The first of many open source contributions we'll be making as a company this year. It's going to be transformative for the downstream industry to have an AI team of this calibre building in the sector, ensuring that the enduring necessity of petrochemical growth can be delivered sustainably - kashmirintelligence.com #AI #OpenSource #ProcessEngineering #ChemicalEngineering #BioChemicalEngineering #Research #Innovation #OilandGas #Refining #PetroChemical

9

🚀 How many of you are using Perplexity AI? Aravind Srinivas, CEO @ Perflexity AI, says Perplexity has been ranked the #1 AI chatbot in a The Wall Street Journal survey, outpacing ChatGPT and Google Gemini! 🚀 Meanwhile, Microsoft Copilot took the last spot. 🥴 #ai #chatbots #technews #innovation #perplexityai #chatgpt #gemini #microsoftcopilot #aifuture #wsj #techtrends #stayupdated

5

🚀 The need for India-focused GenAI platforms has led to a host of indigenous developments. Let's take a look at 8 Made-in-India LLMs that are shaping the country's future. 🔸 Recognizing the socio-cultural gap in existing systems, entrepreneurs and developers from India are working on a GenAI ecosystem tailored for Indian consumption, yielding better contextual outputs across all formats. 🔸 A handful of homegrown conglomerates, such as Reliance (Jio), Tata Consultancy Services, Infosys, and Mahindra & Mahindra, are already working on a slew of GenAI projects. 🔸 Indigenous developments like BharatGPT, Ola Krutrim, and Project Indus (a Tech Mahindra venture including 40 different Indic languages), are addressing the language challenges pan-Indian users face in real life and now in the AI world. According to Inc42 data, India’s GenAI market is expected to grow exponentially in the next few years, surpassing $17 Bn by 2030 from $1.1 Bn in 2023, growing at a CAGR of 48%. With BharatGPT and its ilk taking charge and cementing their positions in India’s conversational AI landscape, India is well-prepared for a new era of digital empowerment where language barriers are dissolved and access to information is democratized. 🌟 Thoughts on this trend: Cultural and Linguistic Relevance: These platforms address India's diverse socio-cultural and linguistic nuances, ensuring inclusivity and effectiveness. Local Innovations and Entrepreneurship: Investments by companies like Reliance, TCS, Infosys, and Mahindra & Mahindra highlight India's growing tech ecosystem and global AI positioning. Economic Growth and Opportunities: The rapid expansion of the GenAI market signifies substantial economic opportunities, leading to job creation and skill development. Digital Empowerment and Accessibility: By providing AI-driven solutions in multiple Indian languages, these platforms enhance digital literacy and inclusivity. Addressing Real-World Challenges: Indigenous AI platforms deliver solutions closely aligned with the needs and preferences of the local population, improving customer service, education, and healthcare support. The emergence of India-focused GenAI platforms is a promising development driving technological, economic, and social progress in the country. Let's look forward to a digitally empowered and inclusive future! 🚀🌐 #GenAI #AI #India #DigitalTransformation #Innovation #Technology #BharatGPT #OlaKrutrim #ProjectIndus #DigitalEmpowerment #FutureTech #AIinIndia

28

3 Comments

Can LLMs continue their historical rise? NVIDIA releases earnings tomorrow. Last week, it became public that OpenAI's Orion is 'less better' than previous updates. The question the prevails: "is AI worth the hype?" This poster writes that AI "can never work in production.” Undoubtedly, AI has changed how people work. Enough to justify valuations? I’m not a PE analyst. When was the last time you saw this level of penetration for something meaningful not TikTok? Either way, let’s dive into it. 1.25b knowledge workers in the world, but let’s focus on 93 million US based knowledge workers and estimate that they save 1-2h per day using AI for basic functions. At an average wage of $27 per hour, that’s 3.7b per day. With 250 work days per year, that’s $941b per year in economic impact. The methodology isn’t perfect, but it should be within an order of magnitude. This doesn't count improvements in efficiency. A product doesn’t have to be perfect to change the economy and the world. Improving efficiency by 10% for 10% of the globe will impact the world exponentially.

2

I participated in a “SMART MANUFACTURING” panel with Gayathri Radhakrishnan, Rohini Chakravarthy and Abhijit Supekar (moderator) last week at TIECon 2024. The coolest part of the panel was 3 Women VCs talking about a topic that has not traditionally been discussed by women. I never imagined I’d see this in my lifetime. Things have indeed changed! We had a very lively discussion of whether we can expect lights out factory in 5-10 years. Automation is happening in all stages but full automation manufacturing is tough because it requires human like adaptability. It is not converting text into instruction or code - it is about precision. Robotics + AI is making this happen and we will get there. AI is going to bring the cost down for complete solution for example: there are robotic arms that can do more tasks and learn like a human and not just some specific tasks. Also check out our portfolio companies Micropsi Industries and RGo Robotics that are making it happen in this space. Thanks to Anshu Bahadur Hrishikesh Sathawane Nilesh Dixit for organizing the panel and Anita Manwani and Surbhi Kaul for making this happen! #TiECon2024 #AIUbiquity

95

6 Comments

Scino360 uses a Framework that combines Finite Element Analysis and Data Assimilation to simulate how a business or founder's product vision will behave under specified market conditions. This occurs in 3 main steps: Step 1: Business Model Design-Discretization The founder's product/business vision is discretized or broken down into a series of variables that fall under the 5 main categories of a Business Model: the Market Segment (S), the Value Proposition or Transformation (T), the Pricing Model (P), the Solution Mechanism (M) and Distribution Channels (D). Step 2: Customer Development The variables are then used as building blocks to constitute a Dynamic Video Sales Letter (VSL) needed to search for a Repeatable and Scalable Sales Process (Test for Desirability). The VSL turns your product features into a compelling story to build emotional connection with the market (Leads). We refer to the asset as dynamic because it is updated over time based on interactions and feedback from the market. Using Statistical Design of Experiments, the Video Sales Letter (VSL) is then broken down into a series of short videos (experimental runs), each of which constitutes a Hook, Story and Call-to-Action. Following syndication of these different videos into the marketing and Sales channels,  the founder can then identify optimal versions of the initial business model variables that optimizes for demo bookings and update the VSL towards product market fit accordingly. Step 3: Agile Engineering The product features presented in the VSL are used to develop a Business Case needed to search for the minimal set of product features that solves the customers problems and results in a reasonable Gross Margin. The Business Case maps low-level engineering features to a financial value to build logical connection with the market (prospects). Following Demo Calls with the prospect and presentation of the business case, the founder can identify the technical product features that solves the customers problem (Test for Feasibility) and align with their budget or specific financial situation while resulting in a profitable unit case (Test for Viability). In summary Scino360’s Scientific Model to product market fit helps tech founders achieve the following: 1.     Validate Business Model Designs: Create virtual models of the founder’s vision to test for various scenarios (Traffic + Sales channels and Market Need) 2.     Optimize Business Model Design: Identify potential problems in a business model design, such as undesirable pain points or product features 3.     Reduce the need for technology prototypes: Simulate product features using a Business Case Model to reduce the need for building prototypes. If your a deeptech founder, looking to find product market fit, get early adopters and agile build an MVP, dm to book a discovery call.

9

LinkedIn team (Juan Pablo Bottaro Karthik Ramgopal) shared a helpful deep dive into their deployment of an LLM-powered job search assistant. Highly recommend this read! Few highlights:  💡 Relatively easy to setup RAG-based pipeline. Achieved 80% of the basic experience in 1 month, then 4 more months to reach 95% mark. 💡 Evaluating the quality of application outputs is hard. I hear this from many enterprises developing LLM applications. (check out Okareo by Matthew Wyman Boris Selitser) 💡 Balancing latency with quality is important. While certain techniques like Chain of Thought could improve accuracy and reduce hallucinations, it will increase response times, which can affect user experience. https://lnkd.in/gtGGwf5P #largelanguagemodels #RAG #EBR #LLM #ml #ai #hallucinations #evaluations #

4

Enjoyable. As a Product Manager, you should obsess over that concept.   Yesterday, I chatted with my friend Abhishek Sinha, a senior product leader at NVIDIA, and ex Cohere and Amazon Web Services (AWS).   In front of 40+ Surge founders, we went deep into all things product management. Some insights worth sharing:   👉 “What is a developer?” - it’s important to go deep and specific on your ICP. For example, you can’t just say that your ICP is a developer. What is that? Frontend? Back-end? Data? In SME? Or Enterprise? There are 20+ different type of developers, each with different needs and motivations. Be specific. 👉 All about motivations- definition of personas and segments should not be done based on superficial demographics. Rather, understand the WHY? Understand the motivations, deeply. Why are they using your product? What drives them? Group them based on the WHY. 👉 60% rule. Product Management is not just about drafting PRD’s and project managing build & launch. It’s all about customers. Working backwards. Those motivations? You can only understand them when you spend at least 60% of you time speaking with customers. Anything less, and you’ll miss stuff. 👉 Storytelling is everything - a product can only succeed when the narrative is strong. Communication is key and product marketing should be joined at the hip and part of all key customer conversations. Again, work backwards from the customer and develop clear messaging and value prop, optimized for your different channels.   Which brings us back to enjoyable. That value proposition? It can only be about 3 things: faster, cheaper, more enjoyable. That last one is not the same as ‘easy-to-use’. Some products are not easy, but so unique, so powerful, so rich that they can be hugely enjoyable.   So aim for that. Aim for enjoyable. cc. Peak XV Partners Anuj Sahai Muharrem Derinkök Jon Woods Pieter-Paul Walraven Jascha Zittel Elliott Gibb Nellie Wartoft James Han Carla Penn-Kahn Nitish Tiwari Ajain Vivek

149

13 Comments

Power of #Gemini #Google #DeepMind Experience the power of Gemini: IIT math question solved in seconds with detailed guidance. Unlock your full potential with AI-enhanced learning. I have only uploaded the image of the question and didn't share any options , the prompt was " Find the value of n " Here's how to solve the problem: 1. Focal Lengths of Mirrors: The focal length of a concave mirror is half its radius of curvature. Therefore: Focal length of M₁ (f₁) = -10 cm (negative for concave mirrors) Focal length of M₂ (f₂) = -12 cm 2. Object Distance for M₂: The object S is at the midpoint between L and M₂, so the distance between S and M₂ is 10 cm. 3. Image Formation by M₂: Using the mirror formula: 1/v + 1/u = 1/f u (object distance) = -10 cm f (focal length) = -12 cm Solving for v (image distance), we get v = -60 cm. The negative sign indicates the image is formed on the same side as the object, which is expected for a concave mirror. 4. Image as Object for the Lens: The image formed by M₂ acts as the object for the lens L. The distance between this image and the lens is the distance between M₁ and M₂ minus the image distance: 20 cm - (-60 cm) = 80 cm. 5. Image Formation by the Lens: Again, using the lens formula: 1/v - 1/u = 1/f u (object distance) = -80 cm (negative as it's on the same side as the incident light) f (focal length) = 10 cm Solving for v (image distance), we get v = 80/9 cm. This image is formed on the right side of the lens. 6. Final Image Coinciding with S: The final image formed by the lens must coincide with the object S for the condition given in the problem. This means the image formed by the lens must be 10 cm to the left of the lens (where S is located). 7. Image Distance for M₁: To achieve this, the image formed by the lens should act as a virtual object for M₁ at a distance of 10 cm + 80/9 cm = 170/9 cm. 8. Object Distance for M₁: Applying the mirror formula for M₁: 1/v + 1/u = 1/f v (image distance) = -n/7 cm (negative as it's a virtual object) f (focal length) = -10 cm u (object distance) = -170/9 cm 9. Solving for n: Plugging in the values and solving for n, we get: 1/(-n/7) + 1/(-170/9) = 1/(-10) Simplifying and solving, we find: n = 34 Therefore, the value of n is 34.

12

We had a full house in a fireside chat with Arvind Jain, Glean's CEO and co-founder, at Stanford University Graduate School of Business! It was an honor to talk with him about building successful companies in the dynamic #AI era. He shared invaluable lessons from his remarkable journey: building Google search infra, founding Rubrik (which went public last month!), and founding Glean, recently nominated by Forbes as one of the Top 50 AI companies! 🎯 Start with the problem, not the solution Arvind is very bullish on Glean's mission to solve app fragmentation and workplace knowledge. He saw his customers at Rubrik struggle with it and took the time to validate it. People tend to undermine the importance of a problem and get too excited about the solution. However, if the solution solves the wrong problem, users will not adopt it. 🤝 Validate hypotheses with people you don't know Friends want to support and might avoid critical feedback. Arvind and the team spent the first year of Glean sending cold LinkedIn reach-outs to potential customers to validate the problem and the solution. While he could easily leverage his network, he wanted unbiased feedback from people who didn't know him. 🎧 When it's noisy outside - stay focused on your customers. That's your moat. With every week's announcements of AI innovations, Glean's Northstar is to be customer-driven and leverage new technologies where there's a fit. Even though they are busy with the growth struggles of a Series D company, Glean teams are working on 70 % of core work : 30% of experimental projects, to stay ahead of the curve. 📈 Companies should plan how to embrace AI internally As a 2nd-time founder, Arvind is looking for new opportunities to use AI at Glean and improve #productivity, beyond using Glean Assistant. With so much innovation, customers are very excited about AI demos but get disappointed when they try to use new tools for their specific use cases. It will take a (short) while before we see a significant impact on productivity.

187

10 Comments

It has been a while since I posted about GLIA Ecosystems I spent 30 years as a lead producer at a bunch of investment banks and boutiques. Successful at times (a couple of exits). Failure at others (a couple of closures). I was a first generation gamer and loved technology. I was an early adopter of Bloomberg back in the days when knowing more fast codes and what the latest product roll outs where made you smarter than the mass. This was back in the days when information was slow, prices were OTC for size, when risk capital was dominated by a handful of banks and when research had price impact. That has all gone away. The game has changed but still retail carries on thinking Trading Views and their ilk (25 mln users looking at charts and back tested technical strategies) can help. Charts, prices, news aggregation and price back testers are nowhere near to a competitive tool kit but it is all most platform users, and market participants, have access to! This at a time when Citadel, Renaissance and their cohort are using weapons of mass destruction to scalp the markets. Guess where all the money is being made? As Robinhood get paid for their “odd lot” order flow and Citadel ride roughshod over it. 90% of trading accounts lose money. 25% of US market flows are retail. 70% of the Western world is financially illiterate. Einstein, or someone close, said “Insanity is repeating the same thing over and over and expecting a different outcome”. That is every data platform user in the markets today, it is the vast bulk of institutional brokers (minus their flow); it is the market at large. We are making leap into the future - a place of differentiation - occupying a space intended to close the gap on the winners; light years (well a good 30 years) ahead of the mass utility offerings. A space where we hope financial success can live. A fund management game with a twist and $1 a day financial intelligence for the mass market. We look forward to sharing much more soon it has been a journey. At long last there is a realisation that mediocre systems breed mediocre outcomes. The future is intelligent; knowledge is curated, content is relevant, and users are at the centre of their digital universe. #learntoearn #joiningthedots #knowingtheknowable #knowledgetoact Peter Bentley Peter O' Toole Adam Toms Matt Barrett Mike Hammond Jon Chase Matthew Fletcher Aba Schubert Ricardo Santos Silva Alf Long Matthew Steiner Mark Gamble Noah Liot Marco Merighi Ameet Patel

67

8 Comments

Attention Product Managers! 🚀 In today's fast-paced AI world, maturing your product leadership is crucial. McKinsey's latest podcast with Inbal Shani dives deep into how AI is transforming product management. 🎙️ It's packed with valuable insights for every PM looking to lead effectively in the age of AI. Here's what you'll learn: 🤝 Foster better collaboration with other teams. 💡 Make AI add real value to your products. ⚖️ Balance new technology with user needs. 🎯 Keep projects focused on results, not hype. 🌱 Lead your team effectively through big changes. Don't miss it! Check it out here: https://lnkd.in/gmSQQtQR #ProductManagement #AI #Leadership #TechTrends #McKinsey #ProductManagers #AILeadership #Innovation

3

H/T to Pieter Kemps and Abhishek Sinha. Great recap below on what great product managers and even more importantly, great product companies focus on every hour, every day, week in and week out. I really believe and have seen the proof points on flywheels that propel you from good to great include these principles: deep and first hand insights into customer motivations, 60 pct rule, storytelling and comms, and designing in faster, cheaper, and more enjoyable… Greta piece!

2

With the recent advancements in AI by OpenAI / Google and the multi-modal LLMs like ChatGPT 4o coming up across the board, education seems to be poised for a very interesting turn, truly! Did you notice the recent video by Sal Khan from Khan Academy on how OpenAI's latest model would be able to hand-hold learners, step by step, using socratic method to get them to learn how to get better at theory and practice under a monitored environment while getting real-time best quality feedback! The mass-level personilized inputs, in the language the users may understand, may just be possible for the first time ever at scale and can help transform learning outcomes per hour being spent by learners The following improvements seem imminent to come: ✅ Improved Access to Quality Education: Multimodal LLMs can help bridge the gap between urban and rural education by providing high-quality educational resources and personalized learning experiences, regardless of location. ✅ Massive Availability of Data: With such multimodal elements, this can make large scale data available for educators/teachers and offer personilised help to students/ suggest courses best suited for students so that people can make better career choices best evaluated and suggested by AI basis their skills! ✅ Support for Multilingual Education: With India's linguistic diversity, these models can support learning in multiple languages, making education more inclusive and accessible to students who speak different mother tongues. ✅ Enhanced Teacher Capabilities: By automating routine tasks and providing real-time assistance, teachers can focus more on personalized instruction and student support, enhancing the overall teaching effectiveness and boosting learning for each student! ✅ Scalable Solutions for Large Classrooms: In a country with a large student population, like India, multimodal LLMs can offer scalable solutions that ensure each student receives attention and support, even in large classrooms and also post class 1:1 doubt solving during homework with personalised guidance that can ensure ✅ Cost-Effective Learning Resources: These technologies can create affordable educational content, reducing the reliance on expensive textbooks and resources, making education more affordable for all. While the tech would be able to solve for, if we are able to leverage the motivation factor and consistency element for learners, we would truly transform the skill level across the population and get massive boost. While doing all of this however, data privacy on learner level would also be a very key aspect! What are your thoughts on increasing use of #ai in #education?

63

8 Comments

Spot on analysis by Sonya Huang and team. Application-layer technology is what will drive large scale adoption of AI. Especially, in country like India where the need for personalization is challenged by vast language diversity and cultural background. As LLMs become powerful in reasoning, it will only make the case of applied AI stronger and expectations even higher. This article has personally added to the inspiration for us at dubpro.ai to build systems that interact and learn from human feedback. By effectively combining a workforce with our AI systems, we have scaled video dubbing across difficult courses and technically complex subjects and dropped cost per hour from $500 to under $40. Agentic AI is working! Very nice read. Source: https://lnkd.in/gqTNZdpJ

3

Evaluating your LLM is very challenging due to various issues such as models’ sensitivity to evaluation set up, cross-method comparisons not being apples-to-apples, and lack of reproducibility and transparency. Luckily these researchers have spent at least three years learning from these obstacles and formulated solutions that will help you skip the line and launch high performing LLM systems. To that end, they present a paper that goes over the common challenges in detail, they provide best practices for reducing the impact of such challenges, and they introduce a new open source library, Language Model Evaluation Harness (lm-eval) for independent, reproducible and extensible evaluation of your LLMs. Do check it out! #genai #technology #artificialintelligence Click #linkedangle and follow for more content Set the 🔔 notification on my page, don’t miss a post!

151

8 Comments