Elise Qian Huang

"Build the event you wish existed" That's what we at Monadical did last week at #NYTW. We wanted an AI Engineers discussion for Engineers. There's always a lot of events out there for VCs to network, or for startups to learn about G2M, but very little on lessons learned from actual engineers in the field. So that's the event we hosted. Big thank yous to our awesome panel, Roy Pereira, Ben Cohen and Corey J. Gallon. Here's the key takeaways and the AI tools we're using. 🚀 All three panelists independently called AI Agents the most transformative LLM application they’ve used. They specifically called out Claude 3.5 Sonnet for its accuracy and reliability. 🪨 We dug into how LLMs are “jagged”, not general. They can be shockingly good at some tasks and completely fail at others. Everyone agreed: good evaluations are critical (and hard.) 🧪 Corey noted how public benchmarks and reality are two different things. Most public evals are saturated or gamed. ♊ Ben emphasized that AI projects are actually two projects: building the tool and building the evaluation process. 🧱 We explored how falling dev costs may impact startup defensibility and labor demand. Roy shared that founders are already shifting strategies in response. ⚒️ In a world of daily AI launches, the panel discussed how they decide what’s worth attention, and what’s just noise. They called out tools like Goose, Aider, Claude Code, and Monadical’s own Cubbi, which helps run agentic workflows safely in dev environments. (links in the comments). CTA: What would you want to hear in an AI Applied Engineering talk you attended?

33

5 Comments

With RL techniques like GRPO, it’s become easy to fine-tune small models to outperform foundation models on vertical tasks. We’re open sourcing Osmosis-Apply-1.7B: a small model that merges code (similar to Cursor’s instant apply) better than foundation models. Links in comments! Using foundation models for high specificity, low complexity tasks like code merging can be overkill - instead, it’s better to use a specialized model instead for higher performance and latency. We fine-tuned the model on 100K real-world code commits where the model needs to merge a code edit snippet into an original code snippet. On the test dataset (subset of CommitPackFT), foundation model performance ranged between 0.77-0.93 reward score. In comparison, Osmosis-Apply-1.7B achieved a 0.98 reward score - while also being 3X-5X more affordable than the next cheaper model AND ~10X faster! Why this matters: the best AI agents use multiple models to ensure task specialization and reliability at scale - on top of cost and latency benefits as well. Links to the model and full write-up (with reward function, hyperparameters, etc.) are in the comments - reach out if you’re interested in using reinforcement learning!

240

18 Comments

Last year, we invested in Cognichip: what we expect to be a generational semiconductor company by making chip design as simple and accessible as vibe coding. They came out of stealth today with a vision for how we will design chips in the future. Today, building chips is nowhere nearly as simple as uttering these commands. In fact, this very task requires going through thousands of pages of documents and hundreds of hours of experienced engineer time. It gets worse with every tech node. Chip design is tedious, which is why chips cost a fortune to design, and take forever to bring to market, resulting in only a handful of (giant) chip companies. Faraj Aalaei, who personally built, grew, and took several chip companies public over the past 3 decades, founded Cognichip with Simon Sabato and Ehsan Kamalinejad to disrupt this status quo: https://lnkd.in/gAkM_k6Q He sounded the alarm a decade ago, and as a dear friend of Lux, he was kind enough to share his vision that became Cognichip: Cognichip has attracted Olympiad-medalist mathematicians and veteran chip designers to make chip design as accessible as coding apps. We expect Cognichip to enable a future where there are as many chip companies as there are software companies, rapidly accelerating compute. We are excited to embark on this journey with the fabulous team Faraj has assembled, along with our friends at Mayfield and FPV Ventures

121

11 Comments

I’ve been working on the theory of why small, shippable factories are the perfect inversion to the centralized prefab factory model. It all collapsed into one line of math (credit to Gilles Retsin): Factory Efficiency = Usage × (Throughput × Product Value ) / ( CapEx + Deployment OpEx) The idea is that the closer you can drive Usage to 100 % while keeping the denominator tiny, the faster the flywheel spins.  A shippable microfactory tackles this by primarily focusing on a high Usage factor. Because the microfactory can be packed up afterwards and reused, its CapEx gets amortized over multiple projects, effectively raising its overall Usage across the year. If one project alone doesn’t fully occupy the factory, it can simply roll over to another job to maintain high Usage. Its CapEx is often an order of magnitude lower than a centralized factory (< $1M vs. $10-50M min), so the breakeven throughput is more achievable on a small pipeline of work. The real breakthrough here is portability. There is near-zero stranded capital and almost no idle time. To flesh this theory out, I wrote 3000 words on the topic for Brad Hargreaves and Thesis Driven. The article breaks down: - why “fixed factories, shipped goods” is being inverted to shipped factories, fixed goods - what a sub‑$1 M robotic cell does to CapEx per home versus a £45 M off‑site plant - early production/cost data on a shippable microfactory from Mollie Claypool, Gilles Retsin, Sam Baker and the Automated Architecture (AUAR) team - the hybrid reality of robots tackling the heavy structural components and human crews assembling faster and cheaper - technical constraints to pull this model off and future considerations on the optimal business model (own vs. rent) Major shoutout to Gilles Retsin for helping me pull this together. https://lnkd.in/eRTnbZPa

54

2 Comments

AI compounds improvements. In August we continued to deploy AI to build AI for investment alpha. August AI product releases from Decasonic include: 💪 Today, we stand at 125 AI teammates to 4 human teammates, a ratio of 31:1 🤖 We shipped +45 agents, 56% MoM growth, above plan ⚖️ We launched our AI Due Diligence Team, comprised of 35 agents performing human quality due diligence at 30x+ speed 📲 We integrated our 10 AI applications in an AI Operating System, unifying how they orchestrate within our workflows 📊 We added +10 agents in our Data Team, focused on on-chain, sentiment and technical market data 🌃 Kudos for the team's hard work, 9-9-7, nights and weekends on top of their day jobs, to ship world class AI. We now have AI that works 365-24-7 More to come...

38

4 Comments

As AI workloads eat up global computing supply, DRAM prices are surging like never before—creating major challenges for teams trying to plan out their data center spend. One of my portfolio companies, Mext, is addressing the largest cost component in the datacenter: server memory (DRAM). From being on the board of Redis for the last twelve years, I’ve learned a lot about the various aspects of this problem and approaches to solving them. Mext found a breakthrough, using new AI techniques, for dramatically reducing the amount of server-DRAM required to run applications, all while maintaining performance. It intelligently manages the server’s memory, keeping hot pages (i.e., those in use by applications) in fast DRAM and offloading cold pages (i.e., those less used) to a much less expensive memory tier (e.g., NVMe Flash). Key to the approach is ensuring cold memory pages that are about to be accessed by an application are back in DRAM before the application needs them or notices that they were gone. This is done without modifications to the application or the OS – so it can run in the cloud or on-premise. Mext thus allows applications to either run using less DRAM or keep their DRAM footprint but do more with it. They're hosting a webinar along with Fred Weber (former CTO of AMD) on Jan 22nd at 10am pacific time. You can register here: https://lnkd.in/gXe9GCeE

153

3 Comments

Move over, Silicon Valley, Seoul’s hardware and manufacturing mastery is rewriting the DeepTech playbook. I saw it firsthand at the 아산나눔재단 (Asan Nanum Foundation) event in Seoul last week, thanks to Chris Lee / Kingsley Ventures and the brilliant Chris Yeh of Blitzscaling Ventures who is always a blast to work with! If you think building software is hard, try perfecting hardware… and Korea’s got it down to a science. All jokes aside, Thank you to 아산나눔재단 (Asan Nanum Foundation) and Kingsley Ventures for putting together this great talk on the pros and cons of Silicon Valley vs Seoul DeepTech VC.

58

7 Comments