Shilpi Agarwal

Satya Nadella sparked a new (or reignited) debate about death of apps as a result of AI agents. While he makes a great point on the Agentic AI architecture and how traditional architectures will be obsolete, wondering if front-end apps will go away. Often, tools like Excel or well-designed user interfaces help me identify the right questions to ask. Can AI alone replace the value of that interactive discovery process? What do you think? #ArtificialIntelligence #AI #AgenticAI #AIApplications #FutureOfApps https://lnkd.in/g8UKWQmf

8

1 Comment

A couple of recent college graduates from India against AI finetuning giants like Anyscale and Run:ai? I met Arko C at a NASSCOM event last October. He was building Xylem AI (YC S24) , a platform that allows you to train and deploy LLMs in production. So naturally most Indian VCs rejected him. Because how could a team of three recent college grads compete with the likes SV startups like Anyscale that have raised hundreds of millions of dollars, have teams of experienced AI researchers and a stash of H100 GPUs? Arko hustled and closed two Fortune 500 companies for paid contracts. I personally know unicorns that haven't managed to do this. He then got into YC in the last round in May. There's a lot to still be proved out, but Arko exemplifies how valuable grit is in building startups. I’m doing a webinar with him tomorrow on challenges of using LLMs in production. Link to register in the comments. Disclaimer: I'm a small investor in Xylem.

153

5 Comments

Effortless Integration for Smarter Insights: CloudEagle.ai Meets Bitbucket Integrating CloudEagle.ai with Bitbucket takes just minutes and gives you: ✔ Simplified App Tracking: No more manual reports, get the app usage data you need, when you need it. ✔ Centralized Control: Manage your Bitbucket security from the CloudEagle dashboard. ✔ Proactive Monitoring: Catch unusual login activity before it becomes a problem. Ready for a smarter, more secure development workflow? Book a demo today - https://lnkd.in/gXAhChrt #saas #saasmanagement #toolstuesday Atlassian #jira #toolstuesday

5

NVIDIA has introduced something amazing—Project DIGITS, which could bring supercomputers into our homes. This means we might soon be able to use powerful AI tools, do advanced research, or create incredible things right from our own desks. It makes you wonder: what’s next? Could we soon see AI in every device we use, making our daily lives even easier and more connected? The possibilities are exciting, and the future looks full of opportunities.

4

Hey everyone, I am an investor and advisor to Katanemo.... They are reinventing the infrastructure primitives for GenAI and launched Arch - https://lnkd.in/gCrbbkEV - an intelligent prompt gateway. Give the project a look on Github, and star the project if it resonates

88

3 Comments

No one cares that it was a hack job. And, it is going to get worse because you were sleeping while they successfully sorted themselves out on their own. Inefficiency makes no difference to the average user as long as it gets done before they notice. The software community has written far less efficient code as we have gotten more capable hardware. You can easily execute a clumsy loop few times before the user can react. There is no longer a need to optimize any program until it is being run at an extreme scale. As we have more fully autonomous hardware systems we are also likely to have less efficient systems. Uber is optimized to get a car to you as quickly as possible at the expense of driver deadhead ratio. You want a car quickly and saving a few dollars through more efficient routing isn't worth the wait. You wouldn't mind traffic so much if you could comfortably sit in the back of the car relaxing. You would have very little incentive to travel at off peak times if you could drink a cup of coffee and read a book on your way there.

17

1 Comment

Latency is important in #WebRTC communication With LLMs, Generative AI and Agentic AI? It is even more important As an industry, we need to figure out how to shave more milliseconds in our processing pipeline Why? Because the LLM piece, along with STT and TTS take TIME⏳ Time we don’t necessarily have This is going to be a big part of my upcoming workshop on #GenAI & WebRTC Register now to reserve your spot! Link in comment

1

1 Comment

The Domino Effect of a Keystone Habit "It wasn’t the trip to Cairo that had caused the shift, scientists were convinced, or the divorce or desert trek. It was that Lisa had focused on changing just one habit—smoking—at first. Everyone in the study had gone through a similar process. By focusing on one pattern—what is known as a “keystone habit”—Lisa had taught herself how to reprogram the other routines in her life, as well. ...'All our life, so far as it has definite form, is but a mass of habits,' William James wrote in 1892. Most of the choices we make each day may feel like the products of well-considered decision making, but they’re not. They’re habits. And though each habit means relatively little on its own, over time, the meals we order, what we say to our kids each night, whether we save or spend, how often we exercise, the empathy we automatically show others, and the way we organize our thoughts and work routines have enormous impacts on our health, productivity, financial security, and happiness. One paper published by a Duke University researcher in 2006 found that more than 40 percent of the actions people performed each day weren’t actual decisions, but habits." ― Charles Duhigg, The Power of Habit: Why We Do What We Do in Life and Business https://lnkd.in/dAvs5PiD #habits #continuousimprovement #growth

8

7 Comments

TSMC | founder has a great story. As with the world’s most valuable tech companies they were founded in dorm rooms, garages and diners by entrepreneurs who were remarkably young. Bill Gates was 19. Steve Jobs was 21. Jeff Bezos and Jensen Huang were 30. Morris Chang however was 55 …. when he founded Taiwan Semiconductor Manufacturing, now one of the world's largest manufacturers of semiconductors. Chang worked for a quarter-century at Texas Instruments and then joined the Taiwan research body that spun off TSMC. TSMC customers include Apple, Nvidia and Qualcomm. Now 92 - his most known quote is as follows. “ ….. Without strategy, execution is aimless. Strategy without execution is useless ….. “

14

Turing award winner Yann Lecun views on AI researchers being awarded the Physics and Chemistry 2024 Nobel prizes. In short, he says that he is excited that Nober prizes go to people working in neural nets. However, Hopfield nets and the Boltzmann machine are interesting models for conceptual study, but nobody uses them anymore, they are practically useless. He also implies that the AlphaFold work does not include contributions to fundamental ideas in AI. #nobelprizes #ai #computerscience #physics #chemistry

3

The keynote speech by Trevor Cai from #OpenAI at #HotChips 2024 was interesting. Trevor's presentation reiterated the ongoing hypothesis at OpenAI—that the AI models must continue to grow to achieve superior performance, with no apparent end in sight... He bolstered this argument using a log-log power law plot, which essentially plots the relationship between compute (in FLOPs) and model performance (i.e., loss) on a logarithmic scale. This approach has enabled OpenAI to accurately predict the performance of models like GPT-4 by curve-fitting the smaller models on the plot, thus demonstrating the scalability of this method. Trevor then drew a fascinating parallel to Moore's Law, which famously predicted the doubling of transistors on chips every two years. This prediction held true for decades, fueling massive advancements in semiconductor technology. He suggested that similar underlying principles might sustain the exponential growth of AI models for a significant time, though the mechanisms might evolve. However, Trevor notes that such growth isn't without its challenges. He highlighted several critical issues, including the need for advanced data centers, enhanced reliability, availability, and serviceability (RAS) systems, and more sophisticated power management strategies. The massive scale of synchronized training introduces vulnerabilities like power draw jitters and single-point failures that can cause widespread disruptions. If synchronized training causes so many challenges, I wonder if the research community is focusing on alternative algorithms for LLMs that don't require gradient synchronization after each iteration.? 🤔 💡 Trevor points out that as AI models became more intelligent, the demand for inference has grown significantly. He notes that while training infrastructure can be planned and scaled, inference capacity must be more responsive to actual usage and demands, making planning more challenging and dynamic. The technology and economics of inference will continue to evolve. While the conference is about hot chips, he has no specific recommendations on GPU/accelerator architectures, whether inference could benefit from custom hardware, etc. Overall, it is evident that openAI will continue to grow the models and datasets it trained on for each new release of GPT and will continue to challenge the AI/GPU and networking hardware industry to support this with technological advancements. At some point in the future, more than half the data on the Internet will have originated from the LLM models. At that point, how would the log-log plot look? 🤔 Wouldn’t models trained on their self-generated data collapse, and the losses start to grow? He acknowledges the problem in Q&A and says they are "investigating" techniques to overcome this…. Any thoughts? #AI #LLMs #datacenters #junipernetworks

82

8 Comments

🌟 Unleashing Potential with NVIDIA Grace CPU: Engineered for the Largest Computing Challenges 🚀 As we navigate the complexities of modern AI models, the demand for enhanced computing power and improved energy efficiency is at an all-time high. The NVIDIA Grace™ CPU steps up as a transformative force in this dynamic landscape. 👾 Purpose-Built Excellence: Specifically designed to address the world’s most substantial computing challenges, the Grace CPU is an exceptional Arm® CPU that sets a new standard in performance and efficiency. 🔗 Dual Capability: Grace excels in versatility. It can be seamlessly integrated with GPUs to amplify accelerated computing, or it can operate as a potent and efficient standalone CPU, making it the bedrock for next-generation data centers. 🔄 Flexible Configurations: Adaptability is key. Grace supports a variety of configurations to meet different needs within data centers, ensuring optimal performance across various applications. 🌐 Introducing NVIDIA GB200 NVL2: Elevating every data center into the new era of computing, the GB200 NVL2 platform delivers unrivaled performance for tasks like large language model inference, vector database search, and intricate data processing. Embrace the future of data centers with NVIDIA's cutting-edge technology, crafted to push boundaries and drive innovation. #NVIDIA #GraceCPU #DataCenters #AI #MachineLearning #DeepLearning #TechInnovation #FutureOfWork

11

Evaluating my fine tuned 3B model with LLM judge - many of you have asked so here goes :-). Tl;Dr - 91% accuracy vs 47% not-fine tuned; with only 140 examples and 3 minutes of training (https://lnkd.in/gHw993ZS) I started out with creating a new synthetic test data set that the fine tuned models haven't seen, using the same code from the training set (data here: https://lnkd.in/gcCSdT5e). I used Comet's Opik per Armand's recommendation (you should follow him if you aren't :-)). Then built my evaluations test for (1) JSON schema compliance, (2) distance from reference answer (3) the LLM Judge which scores first on json format, then on compliance with schema, then on entities being detected, then on not detecting more than necessary. I tried this out with a bunch of LLMs that could run on this laptop, as you can see the 3B fine tuned one performed the best, followed by the 1B, then all of the out of the box ones pretty similar and not so great. Code for you to enjoy: https://lnkd.in/guhzdUkf You'll need a Comet account, which is shockingly free. Enjoy :-) Next I'm going to try to beef up the data gen with Distilable and CrewAI, let me know if you find that interesting or have other ideas.

200

19 Comments

Prevention and static rules only goes so far. If you’ve ever had that magical moment with FSD or when you first took a Waymo you need to ask yourself why you’re ok with non-realtime systems that don’t adapt for managing cloud infrastructure. Or maybe you’re printing Gold Master CDs.

4

1 Comment

This is worth a quick listen. Watch beyond the no-task-beneath-me part to how he shares his logic and reason.

7

The best analogy for GenAI is the hammer. 🔨 If you need to nail something down, it is going to be tremendously valuable. For most other tasks, it will be useless at best, destructive at worst. Note that this is true for _any_ technology and this is where expertise and experience really helps: you can buy an expensive toolbox for fixing stuff at home, but if you use a hammer where a cutter is needed, you’re going to have a bad day. #justsaying #ml #ai #genai

11