Adrian Macneil

Little can be added to those statements. Hardware can be one of your best competitive advantages, if done right.

6

2 Comments

Very excited that VitalStage Ventures is a small part of Groq’s new round of funding. We are all-in on AI infrastructure as the picks and shovels enabling this new generation of innovation. With respect to Groq, we particularly love their new hardware + software approach to accelerating AI inference. Just try out Groq.com and you can sample what the fuss is about. Having specialized Language Processing Units (LPUs) for inference enables dramatic performance gains at higher efficiency with less thermal impact. And now by hosting models via GroqCloud, Groq lowers the barriers for teams to be able to leverage many models without having to set up and deploy their own AI infrastructure. #ai #groq #ainative https://lnkd.in/gcEcXfme

44

1 Comment

Today at Predibase we've added fine-tuned GPT-4o-mini to our Fine Tuning Leaderboard, to quantify how it compares against top open source SLMs like Llama-3.1-8B and Solar Pro across 31 diverse tasks. To our surprise, while 4o-mini had the strongest out-of-the-box performance of any base SLM before fine-tuning, the lift from fine-tuning was pretty minimal, ultimately landing in the middle of the pack after fine-tuning. This is consistent with what we've heard from our customers, who very often benchmark against 4o-mini and find it to be worse performing than the best open source SLMs, in addition to being significantly slower and more expensive over time. What should we conclude from this? For me, the takeaway is: don't over-index on out of the box base model performance. Most of the value will come from your data, and using a model + fine-tuning platform like Predibase that can get the most from it. Check out the complete results here: https://lnkd.in/gBPGY6_k

75

6 Comments

 🔥 "How many engineers does it take to replicate Docusign as a software product in 2026?" Sarah Guo dropped this thought-provoking question at last week’s panel. Answers from the eng/product leaders at top companies were eye-opening: Scott Densmore: 10 engineers Romain Huet: 5 engineers Ainsley Escorce-Jones: Just 1 Lee Robinson: 1-2 engineers Say hi to a future where rewriting major software might be simpler than you think. Reach out to me if you are planning to do that! 🙋🏻‍♂️ PS: Take a look at what Gil Arditi is working on at Morph.ai 💡 #software #sdlc #softwaredevelopment #llm #ai #ml #largelanguagemodel #copilot #gpt4

29

13 Comments

To build a multi-agent system, start simple, validate your design, then gradually scale. When I spoke to Chi Wang, the creator of AutoGen, he explained why. Deploying one or two agents at first allows builders to evaluate and refine the core design and interaction patterns before introducing additional complexity. This method also streamlines debugging and optimization because it makes it easier to trace issues back to specific agents. Our full conversation here: https://lnkd.in/gEKmExu4

15

1 Comment

Question for VCs and executives in agile manufacturing: Desktop Metal, the darling of 3D printing, is about to be sold at its early-days valuation. What did we learn from this? Here are a few lessons I picked up: 1. Focus on verticals: It's incredibly tough, if not impossible, to build a general-purpose machine. Manufacturing is full of exceptions, and each application has its unique requirements. Don't spread yourself too thin. Dive deep into the most promising vertical first, then expand to the next. 2. Have technically savy managers: Manufacturing requires both technical know-how and operational experience. Ensure your executives in agile manufacturing companies understand all the technical tradeoffs and can decide what not to do and what to focus on technically. What not do is probably more important that all the things we could do. 3. Hybrid processes win: Don't define a solution for agile manufacturing as a single process like 3D printing. Hybrid approaches are key. A single part often requires multiple operations. That's why in 2018, we envisioned a process-agnostic platform capable of utilizing various methods—be it 3D printing, machining, or Roboforming—through AI to achieve the best results. This vision led to the birth of Machina. 4. Have high conviction but be ready to change: Despite the clear limitations of 3D printing, many investors, including those backing Desktop Metal, advised us to sell out to companies like them till recently. Their strategy didn’t seem to work, and yet, some praised it until just a few months ago. What did you learn? I would love to hear from investor in DM. Note: I am still very bullish on 3D Prining but with a different product and GTM strategy.

120

17 Comments

Great report from Menlo Ventures showing the potential and traction of Enterprise-grade AI. “In 2024, much of the action happened at the application layer. With many architectural design patterns established, app layer companies are leveraging LLMs’ capabilities across domains to unlock new efficiencies and capabilities. Enterprise buyers are seizing the moment, pouring $4.6 billion into generative AI applications in 2024, an almost 8x increase from the $600 million reported last year.” https://lnkd.in/g4j_QCZu

13

1 Comment

Excited to share our 0-1 #DevGTM Playbook for early-stage technical founders building developer tools and infrastructure! We at Speciale Invest had the privilege of working closely with many brilliant technical founders and evaluating many Devtools/Infra Cos. During our in-depth discussions, we discovered a recurring obstacle that many of these technical founders face: despite identifying problem statements and developing cutting-edge technology but crafting a clear and effective GTM strategy often proves to be a significant challenge. Recognizing this gap, we set out to create a comprehensive playbook tailored for early-stage technical builder/founders in their 0-1 journey. My colleague Aryan Pareek and I have spoken to over 50+ #DevTool founders, GTM experts, and #developers from companies like Amnic, Appsmith, DhiWise, Dyte, Esper,Hashnode,Hivel,Keploy 🐰, LambdaTest, Neverinstall, Portkey, Reo.Dev, Requestly (YC W22), ToolJet, Tailcall, Typo, Unstract , videosdk.live, Zenduty and many more. And we consolidated the best examples and learnings into our playbook. We dive deep into various aspects of #developer GTM, including defining your ICP, buyer journeys, messaging for developers, #SEO, content strategy, and more - all tailored for dev founders. Check out the playbook here: https://lnkd.in/g425vuqy Huge thanks to all the #devfounders and #GTMExperts who shared their invaluable insights and experiences with us. Special shoutout to Dev founders and all the amazing developers and GTM experts who gave us their valuable time. 🙏 📣 Special Creds to the devfounders who helped us with this: Sachin Jain Achintya Gupta Navaneeth Padanna Kalathil Vishal Virani Anirudh Murali Shiv Sundar Kshitij Mohan Sudheer Bandaru Venkatesh Radhakrishnan Harini Janakiraman Arjun Kava Narendran Hariparanthaman Sidhant Goyal Madhukar Kumar and many more.... (lol - I've exceeded the maximum mention limit) Special thanks to Freshboost for assisting us with SEO strategies. *Please tag technical founders/builders who would find this useful, and feel free to share your feedback with us over DM* P.S.: It's still a work in progress, and we'd love to hear from more technical founders about your experiences and challenges about your GTM journey. Your feedback will be invaluable in making this resource even better for the broader dev ecosystem. Arjun Rao Vishesh Rajaram #TheDeepTechVC — Build from India for the World

167

15 Comments

The idea is easy. Manufacturing and mass production is the difficult piece of growing any product.

6

What a time to be alive. Yes, LLMs are reaching initial plateaus. But then again, the underlying compute is getting 2-3x cheaper and more energy efficient every year. Our models today, are around 40.000x more compute efficient than comparable numerical models. Already today, we can build with 5M or 10M investment, what used to take 200M+ just in infrastructure before. In one year, we can build the same for 3M. In two years for 1.5M. In three years for 0.75M (or less). In other words, a global, high resolution, high precision world model, that would have taken billions to build just 5 years ago, will be built with a few million of investment in the next 5 years. This is our exponential journey towards climate sentience.

48

4 Comments

I'm concerned about the future of society if everything becomes automated. What will be the role of humans in such a scenario? And what exactly is the mission of automation - to replace human jobs or to enhance human productivity? Ultimately, what is the mission of human society? #automation

4

1 Comment

Lead Your Agrifood Business into the Future with AVA As a CEO, especially of a startup, leading your business via innovation is crucial. At SWARM we started with an optimization platform to solve complex agrifood supply chain problems, added a dashboard to make it easier for clients to interact with the technology, and then introduced Challenge Engineering ™ to let people have a structured and repeatable way of defining problems – so that AI could solve it automatically. What’s next, you ask? Well, that would be AVA, the Agrifood Virtual Advisor, and every day I’m blown away by her capabilities. Powered by OpenAI ChatGPT, but extended with knowledge of agrifood, challenge engineering, and trained on customer’s own (fully private) data, AVA helps you identify and solve key challenges, from logistics to network efficiency. She captures and refines corporate knowledge, offering actionable insights to drive your business forward – and storing your institutional knowledge in a permanent memory. AVA is more than an AI; she’s your consulting partner to improve your organization. Ready to lead your business into the future? Learn more about why some of the world’s biggest agrifood enterprise companies are leaning into AVA at SWARM Engineering. #Leadership #Agrifood #AI #Innovation https://lnkd.in/eP-8DTpQ

16

I've spent the past few months meeting with leaders from the most successful startups in robotics, trying to understand what the next generation of robotics companies will look like. I consolidated my learnings into 3 essays: 1. A Market Map of Sorts 2. The Control Conundrum- The evolution of robotic control 3. Humanoids to the Moon- The over valuation of humanoids and robotic foundation models What I find interesting about the space, is how the hype surrounding humanoids and robotic foundation models, is quite divorced from how robots drive revenue today. I believe in robots that automate processes. I believe in robots that are 99.99999% reliable. I believe in robots that are focused so they can be effective. Think you're building this? Lets talk. Disagree? Lets talk. Check out my twitter for snippets of these essays if that's more your speed. Links in comments.

91

5 Comments

AI's hallucination problem is more than just a glitch - it's a 41% error rate that's holding back real-world applications. Enter knowledge graphs and GraphRAG: powerful tools grounding AI in facts and relationships, making outputs more accurate and explainable. For businesses seeking to harness AI's potential, this could be the key to unlocking trusted, practical solutions. #AIHallucinations #KnowledgeGraphs #AIAccuracy #EnterpriseAI #InnovationInTech What steps is your organization taking to improve AI reliability?

6

3 Comments

Are AI co-pilots a sub-optimal solution? Kind of like the way Elon Musk thinks about Hybrids. Tesla doesn't make hybrids because the goal is fully electric. Hybrids are a detour on the path to fully electric. If GTM teams are facing a scalability and sustainability issue based on human constraints, are Co-pilots just a sub-optimal detour on the way to fully autonomous?

10

1 Comment

Remember Jian Yang’s app, SeeFood, from Silicon Valley? It was a "hot dog detector" that scanned food images and told you whether it was a Hot Dog or Not Hot Dog. As expected, it flopped. But after shutting down SeeFood, Jian Yang’s product was acquired for $4M by the company where he was interviewing. Why? Because they realized the same algorithm could be repurposed to filter inappropriate adult content online—NOW THAT'S A REAL PROBLEM. That’s the power of product positioning: Same technology, different positioning, different results! Over the years, I’ve seen even the best products fail due to poor positioning. So, how do you get it right? Talk to as many people as you can! Use those conversations to build your positioning and narrative. At Sanas, we initially positioned ourselves as an 'accent translation' company. However, after speaking with customers, we realized that we were actually enabling people across the globe to understand each other better. As a result, we repositioned ourselves as a 'speech understanding company.' Since then, we have developed a Noise Cancellation product and are working on Speech Restoration and Speech Enhancement products — all with the spirit of facilitating real-time speech understanding. Positioning your product correctly is as important as building it. Make sure you get it right. Maxim, Shawn, Marty, Ashish, Anant, Monal, Supriya, Andrés, Carya Venture Partners

155

9 Comments

NVIDIA's Jensen said "ChatGPT democratized computing, Llama2 democratized generative AI" ...and now Llama3 is the next level 🔥 It's fascinating to watch Meta's strategic moves. With Llama3, they prioritized training on a massive 15T token dataset to pack all into a lean 70B param model, instead of building a massive model. This allows Llama 3 to match trillion+ parameter models like GPT-4, but at 1/10th the compute, storage and inference costs! 💰 This technique was published by Google DeepMind a few years ago https://lnkd.in/gx7VU8aA Meta is not a dark horse in AI anymore. They might be the top dog. https://lnkd.in/gRRJYuxt #llama2 #llama3 #llm #chatgpt #gpt4 #ai #ml

16

5 Comments

Anthropic just released its new multimodal AI model: Claude 3.5 Sonnet. The benchmarks sure need a refresh as AI models keep improving at breakneck speed. But more importantly, each benchmark represents a real-world application, so it's crucial to steer clear of thin wrappers over any benchmarked capability. #claude #anthropic #chatgpt #gpt #openai

5

1 Comment