Abhishek Sharma

We are excited to back Morphing Machines. As Dr. Ranjani, CTO and co-founder says: “For the first time in India and globally, we have a processor that is truly runtime reconfigurable, catering to emerging workloads.” While Nvidia’s growth and high valuation are driven by robust demand for “infra-compute” in high-performance GPUs, spurred by Gen AI, there is a distinct need for innovative solutions tailored to a new wave of reactive applications. Enter Morphing Machines. It addresses this demand with its unique product designed for next-gen telecom, autonomous vehicles, collaborative robots, avionics, and online data analytics. This product is specifically engineered to handle heterogeneous workloads and diverse tasks with varying performance demands simultaneously, ensuring a high level of scalability. As we move from an institutionally-backed research team at IISC to productization, Golden Sparrow is collaborating with Speciale Invest and others to support Deepak Shapeti, who joined as CEO three years ago after earning his Masters at Stanford. He brings not only technical expertise but also the business acumen necessary to guide the product to commercial success. Venkat Raju Afreed Faizan Anand Rao Ritik Singh Michael Marmor

68

5 Comments

Founding father of Bittensor Jacob Steeves just shared some very profound insights on how the consensus algorithm of the network was designed … and what it points to. 🌱 “We built the API and consensus algorithm to be as general as possible—necessary for the first digital commodity we were evaluating which focused on miners producing information-theoretic value. Pure information is so fuzzy and high-dimensional that we had to abstract the consensus algorithm to work over a weights matrix, not tied to any single digital commodity. Note: this is similar to what makes neural networks powerful: deep learning invented tools that could learn anything, rather than prior algorithms that focused on specific tasks (e.g., detecting edges by coding edge concepts directly). So you're partially correct: Bittensor's blockchain has no ML on it (though the consensus algorithm is heavily influenced by ML; it uses sigmoids and weight matrices). We didn't want to apply crypto to AI or AI applied to crypto—we wanted to be both. Getting esoteric: I believe doing anything in this space will require uncovering the deeper reality to both crypto and ML: ML is the language of minimizing loss curves; crypto is the language which organizes incentive-maximizing individuals and machines. They only fit together because they're expressions of the same thing: This is how nature organizes -- it uses the flow of energy dissipation through distributed systems to organize. We're not ML; we're not crypto; we're trying to leapfrogged both and go where they are pointing.” https://lnkd.in/ggE-MQ-x

21

Alphabet committed another $5 billion recently to Waymo's self-driving cars. A renewed vow of confidence. Per my post yesterday, fast growing additional compute and better AI architectures are changing self driving cars rapidly. Notably, some traditional car manufacturers pulled out investments in full self-driving cars. It didn't think the tech was going to be ready for decades to come, if ever. I think it may be a mistake given AI and compute grow exponentially, not linearly. As an example, it may take 20-40 years to achieve 1% of progress, then just 5-10 years for the remaining 99% (instead of 1,000s of years per our intuition) link in comments #AI #Transformers #waymo

2

1 Comment

Interesting comparison of human-based ride hailing services (UberX, Lyft) vs autonomous-driving one from Waymo. TLDR: Waymo is more expensive and a lot slower to get you to your destination. And this is in the city where Waymo vehicles and software are probably most optimized for. This isn't to bash on Waymo and other self-driving car platforms, which have made incredible progress. That being said, the problem is not solved and it will take many more billions and many more years to get autonomous driving services like this to be price and performance competitive to current ride hailing services. https://lnkd.in/eb4uZ2Um #autonomouscars #AI

1

172. Software is Feeding the World What are the similarities and differences in the adoption of steam power, electric power tractors, and generative AI solutions within agriculture? This week's newsletter looks at how solving the problem of power drove the adoption of tractors and a dramatic reduction in the US labor force participation in agriculture. I draw on the early history of the first tractors in the late 19th century based on the work done by Neil Dahlstrom. AI and Generative AI are about precision, and the adoption depends on about economically viable precision, as was very eloquently pointed out by Walt Duflock, when he posted about apple picking by advanced.farm Will a broader adoption of AI lead to structural changes, where companies might have to invest more in the training, engagement, and retention of human experts to differentiate themselves, as many farmers still prefer a human component? (Based on McKinsey & Company Global Farmer Insights Survey by Vasanth Ganesan et al) I would love your perspective. Also, if you sign up for an annual subscription in the next 3 weeks, you get a 15% discount. https://lnkd.in/gehyayHZ

14

1 Comment

AI: the bubble that burst What’s wrong with the current unscientific AI. Let’s summarize the WSJ article: * Uncertain return to investors * Unrealistic goals * Not delivering on its promises * Narrow set of practical applications * Many model variations of the same algorithm * Taking a long time, or never, to realize the business case * Spending does not guarantee a return * Pace of improvement slowing, or too many errors * Imperfect data; data being appropriated from others; * Open source options catching up with paywall models * Excessively expensive to run and maintain for meager results * Public starting to realize of its limits: self-driving cars fatal accidents; poor search results; lawsuits by owners of the data; driven by personalities not science; increasingly taking more electrical power from the grid Finally, at the end of the article, WSJ closes with: “The problem is that the current level of investment—in startups and by big companies—seems to be predicated on the idea that AI is going to get so much better, so fast, and be adopted so quickly that its impact on our lives and the economy is hard to comprehend. Mounting evidence suggests that won’t be the case.” Source: The Wall Street Journal https://lnkd.in/gg-7hJjN #artificialIntelligence #ai #machineLearning #dataScience #spe #petroleumEngineering

155

61 Comments

Big tech continues to go for mega acquisition. 2022: Google’s $5.4B acquisition of Mandiant 2023: Microsoft’s $69B acquisition of Activision In 2024, we are seeing deals like Synopsys’s $35 billion acquisition of Ansys Why do tech giants acquire companies? Well yes, some of them sit on large piles of cash. But then they could be giving out dividends. Why invest in acquiring companies? Here’s why: Growth Companies like Amazon, Apple, Google, Meta, Microsoft, and Nvidia are acquiring firms to boost their capabilities and market reach. This helps them eliminate competition and access new technologies and talent. So if you are a business owner looking for a 1B+ exit, assess whether your technology, market penetration, or resources can be of any use to these big names. #entrepreneurship #venturecapital #startup #mergers #acquisitions Serial Entrepreneur & Investor Helping Startups Become Unstoppable – David Hauser

85

3 Comments

Using generative AI, NVIDIA operations built an AI Planner agent, developed on NVIDIA Inference Microservices (NIM). The agent leverages LLM, NeMo Retriever and CuOpt NIM to reduce re-planning time from hours to just seconds. https://nvda.ws/4bIfS2T

18

"So getting back to NVIDIA, I don’t see any force stopping the juggernaut in the next few years, other than a huge decline in the need for AI compute, and that seems unlikely. Any startup competitor could easily be acquired by NVIDIA for tens of billions of dollars and it would barely be material given NVIDIAs market cap. And while the other major chip companies are mounting a challenge, particularly AMD, the demand for GPUs is so great, and will be for the next few years, there is room for massive growth for multiple players. When NVIDIA eventually slows down, I suspect it will come from one of two areas. The first is a surprising shift in market use cases. Given the long development cycles of computer chips, it’s hard to be as responsive to the market demands as software companies can be. And product designers have to choose some level of predictability and reliability in their components and their partners, which is why they favor proven tech over innovative new designs most of the time. But the way NVIDIA benefited because their chips for graphics cards happened to be a good match for new AI workloads - I think something similar could happen to slow them down. Some new popular AI workload will be a better fit for something other than a GPU, but a chip that is already established in the market in other ways. Or secondly, it comes from a company that is building their own chips and decides to get into the broader semiconductor business. This means Apple, Amazon, Facebook, or most likely - Google. But all those companies, even if they can compete technically with NVIDIA, lack the infrastructure to sell and support chips and those who develop them. Building those capabilities will be slow." https://lnkd.in/dFpj5hqv

1

1 Comment

Great insights on the future of robotics shared by Rodney Brooks in this article. A little bit of realism for the hype surrounding LLMs in robotics - realistic expectations are important...

4

Here is another misuse of Artificial Intelligence If the science and physics of the state of the elements (solid, liquid, vapor, and plasma) is already settled and understood, why people insist in using machine learning to rediscover what we alll already know? To me it is a symptom that pseudo-AI companies are pushing researchers to use their software to recreate physics with only data. I have seen this happening in respected oil and gas companies. Usually this occurs when the IT departments take too much control of the data science and machine learning workflows that engineers use every day. My message is this: do not succumb or fall in the temptation of so-called AI shiny software. Know and learn you science and physics first. Keep in mind that marketing or sales people are not in their majority engineers or scientists. #artificialIntelligence #AI #spe #petroleumEngineering Massachusetts Institute of Technology

40

5 Comments

Why I am not surprised “Further down the value chain, away from the glow of Nvidia, lurk signs of discontent. Businesses have cut back on whizzy new AI tools out of concern for hallucinations, cost and data security. “ “First, tech companies need to identify where their hype machine has gone wrong. They didn’t set expectations for AI’s capabilities too high; they framed its use as being too general purpose.” “Framing AI as a general-purpose Swiss Army knife for productivity inevitably leads to paralysis for its end users: Where do you even start with a technology that can do everything?” “AI isn’t yet a jack-of-all-trades but a master of a few. The sooner business leaders realize they can apply it to an array of niches and not for everything, everywhere, all at once, the sooner they can make the technology useful for them.” *** What prompts this analysis by Bloomberg is the meteoric rise of the stock price of NVIDIA. It is producing great concern because it resembles the internet bubble we saw 20 year ago. There is no basis for the 3 trillion USD value of the GPU company. How do we start correcting course: 1. Stop calling data science and machine learning AI. They are not interchangeable terms 2. Treat AI as a science not a mercantilistic goal 3. Don’t buy “AI” solutions for the sake of being trendy and smart. You will get the opposite effect 4. Start slow. Build your foundations with statistics, data science and data quality processes. Examine the need for machine learning algorithms 5. Inform yourself about what AI is. Put your hands on an old good friend: books. 📚 Be extremely selective with your internet choices of reading, mindful of the internet press makes money by your clicks and your data 6. Set realistic goals and establish a real, concrete result if you get hooked with “AI”. If a vendor is too insistent, try a demo or a pilot of the “AI solution” for six months, at no cost 7. Ask the vendor for real world results. Data is not perfect. You may want to try the “AI” product on your data first 8. Move your focus to the data. See if your project or challenge qualifies for machine learning. If your data is imperfect, solve the issue with the measurement processes upstream. No algorithm or “AI” does miracles. It’s all about the data 9. Be specially cautious if the vendor claims to replace your “old” and “slow” physics based models, or sound engineering workflows, with data-driven only, machine learning, or solve-it-all super fast “AI” 10. Digital Transformation is not about switching old toys for new shiny toys; it is about improving efficiencies, solving business problems, make the intricate reproducible. There is no point spending $100 million to obtain a benefit of $1 million 11. Use common sense. Ask your engineers and end users on the “AI” product. Involve them, use the bottom-up approach, and not the power of your title #AI #artificialIntelligence #spe #petroleumEngineering #machineLearning #DigitalTransformation

11

1 Comment

Are you the curious type that enjoys discovering and learning exciting new sub-areas and applications of Machine Learning? Want to get a deeper understanding of how ML is used for physics and engineering? Heard Jensen Huang repeat excitedly "Digital twin", and you couldn't understand what's the big fuss about? Then this awesome YouTube Play list is for you - Physics Informed Machine Learning - great set of video lectures by Steven Brunton. The lectures are engaging and clear, even for people with fairly basic background in ML and Physics. It's still evolving, but there is already a great set of new introductory lectures, as well as curated deep dive lectures by Steve from previous years. This is a somewhat similar course from the CAMLab in ETH - https://lnkd.in/eDK-u8SS, which I watched last year. It's also excellent and covers the key topics, though I think it's a bit less accessible than Steve's. Try both and let me know 😄 I recommend! 👍👌 https://lnkd.in/eMeKWtdx

37

3 Comments

This is an extremely thoughtful and well observed piece Tim Lee has written (thanks John for posting). Well worth reading. My take: Almost all credible autonomy technology companies are using neural networks in (increasing) parts of their architecture and development processes. They have an unmatched ability to learn patterns, recognise complex scenes and generalise at scale. But it is also true that they alone may not (likely will not) get you all the way. Many of the reasons for that are called out neatly in this article. But there are also some omissions. There are some useful hints in the recently passed UK AV Bill that refers to “in-use monitoring” as well as “explanability and transparency”. In essence; in the event of fault, the Authorised Self Driving Entity (ASDE) must be able to discover and explain the fault and then rectify it. This is inherently difficult for neural nets but standard for safety critical system. AI’s routinely make mistakes or experience phantom recall. That might be okay in training data. But it’s also possible these mistakes first appear at run-time when it’s too late and there’s no way to discover it and test for it ahead of time. Though we each have our unique magic star dust, I would wager Oxa’s architecture is more closely aligned with Waymo’s, at least philosophically and from the bedrock of safety and validation. The key word here is “validation”. They, like us, probably use a combination of technologies and techniques that match the right need. It’s likely some of those techniques utilise as many complex specialist algorithms as they do combinations of compound AI’s. Its understandable to assume that geographic restrictions at first blush might seem clunky and slow to scale. The honest answer is the early versions are. However, our bet (and I’m certain Waymo’s bet) is that the speed in which we are innovating to validate new “autonomy ready” geographies will outpace the speed in which black-box neural-net only approaches can be ready for driverless operation (if ever). This is the fastest route to generalised, validated, safe autonomy for everyone. Or what we like to call “Universal Autonomy”.

35

2 Comments

Batch processing for LLM inference activity: "One sign of that is the rise of batch processing. Today, popular consumer AI products like ChatGPT and Perplexity provide users with near instantaneous responses, otherwise known as real-time inference. However, businesses don't always need immediate responses and are often willing to wait hours, days, or even weeks for responses, as long as they don't have to shell out as much money. Founders of cloud and inference providers have told me there's growing pressure from business customers for this kind of flexibility, otherwise known as batch processing." -- Nathaniel Whittemore on "The AI Daily Brief" (with much deserved shout-out to his inspo from Stephanie Palazzolo at The Information) I fall into this bucket as we have a lot of backfilling to do as we're building out the underlying data in our service, and some of that could just as well be done in the wee hours of the night. The new stuff flowing in I want processed ASAFP, but these backfills—I'd mos def take more cost-efficient delayed batch processing. There's also use cases like needing (near) real-time inference for the first inference pass for timeliness / contemporaneousness value, but then a second pass slow-time inference for wider classification exercises over an ever-growing data corpus. I could see us mixing Sonnet & Haiku models in both cases. Just a heads up to Anthropic Mike Krieger as you contemplate this, along with Amazon Web Services (AWS) Jeff Barr Mark Roy Swami Sivasubramanian on the Bedrock side of things...and I don't mean the "batch" mode that doesn't seem to have any real value other than helping AWS manage its own rate limit challenges. (DeepCast is using Bedrock.)

11

2 Comments

Will AI and Aging Demographics Accelerate Robotics Adoption? The robotics market is at an inflection point driven by convergence and advancements in AI. We expect robotics adoption to accelerate in the coming years as factories and warehouses deploy more automation to address America’s shrinking labor force. In the coming decade, Humanoid robots could fill the growing labor gap from our aging population in just about every industry from healthcare, farming, retail to construction and mining. If you are an innovative robotics startup building autonomous systems or enabling tech and applications, as a frontier tech VC, we would love to hear from you. We attended the AI & Robotics Summit recently, sponsored by the Special Competitive Studies Project (SCSP). The purpose of the event was to bring leading experts across industry, government and academia together to discuss and demonstrate the latest tech and highlight that US leadership in robotics is critical to our economic future and national security. #AI #Robotics #Autonomy #Drones #AIRobotics #Autonomous Vehicles #UAS #Unmannedsystems #Industrialrobots #Humanoid

18

8 Comments

Infrastructure advancements are facilitating a new era of DataOps, enhancing data management and changing how we see / use AI applications. Have a look at our latest thinking on the AI infra layer!

10

1 Comment

Large language models (LLMs) are still in their early days, but they're notorious energy hogs, mainly because they spend so much computational power on floating point tensor multiplications. Enter BitEnergy AI from Cambridge, MA—they've come up with a clever algorithm called L-Mul that approximates these heavy-duty multiplications using simple integer addition. This isn't just a neat trick; it has the potential to slash energy costs of element-wise tensor multiplications by up to 95% and dot products by 80%, all while maintaining or even improving precision. Innovations like L-Mul could dramatically reduce the outsized energy requirements of LLMs, making them more sustainable and efficient as they continue to develop. You'll see this trend of faster, more efficient LLMs hitting the market continue to pick up steam over the next few years, leading to cheaper inference costs and a wider range of cost-effective use cases for AI.

3

1 Comment