Tech Thoughts: How to Play the AI Long-Game
Quick recap on what's going through my brain. This week covering: DeepTech strategy, hardware for LLM training in startups, LLM performance, VCs and global living conditions
“Sometimes, you have to ignore your customers”
Jensen Huang, speaking about the early days at NVIDIA when the company trusted in Moore’s Law so strongly that they initially made GPUs that were too expensive and powerful, and that no one would buy.
Must Watch
What I’m Learning
Full Stack Deep Tech - by Ian at Carlos Ventures
Exploring the power of full-stack startups, this post dives into why selling complete products in deep tech outstrips traditional tech sales. With insights from Amazon to SpaceX, it reveals how this approach reshapes industries, offering startups control, market potential, and a transformative path to becoming industry incumbents.
Training great LLMs entirely from ground up in the wilderness as a startup - by Yi Tay
Reka team member shares the gritty details of building infrastructure for large language and multimodal models from scratch. Highlighting the unpredictability and variance in hardware quality from compute providers, it delves into the challenges faced, from unstable clusters to poor I/O systems, emphasizing the hardware lottery in the era of LLMs. Despite these hurdles, the team navigated through with makeshift solutions, emphasizing the importance of resilience and adaptability. The post also touches on the learning curve in coding outside of Google's ecosystem, and the pragmatic yet successful approach to model scaling, underscoring the adventure and hard-earned insights in their startup journey.
It was never about LLM performance - by Technically
Highlights the LLM community's focus on benchmarking model performance, critiquing it as a narrow measure of value. It argues that the real differentiator in the near future will be the user experience built on top of these models, not the underlying technology itself. Using ChatGPT as an example, it demonstrates how OpenAI's enhancements create a superior user experience beyond just the raw output of the model. The post also touches on the challenges open source models face in matching this experience, suggesting that success in LLMs will hinge more on the holistic experience providers offer than on achieving performance parity.
AI startups require new strategies: This time it’s actually different - by A Smart Bear
Unlike previous tech revolutions, AI startups can't rely on traditional dynamics to outmaneuver incumbents. With incumbents eagerly embracing AI, leveraging vast data, and innovating quickly, startups must develop novel strategies to compete. Success now hinges on creating unique value and navigating a landscape where incumbents hold significant advantages.
The Best Venture Firm You’ve Never Heard Of - by The Generalist
Hummingbird Ventures, a venture firm known for its discretion, has achieved remarkable success by focusing on identifying unique founders. With a strategy emphasizing the importance of the entrepreneur and a penchant for contrarian thinking, it has consistently delivered returns above 10x, favoring founders with neurodivergent traits and resilience through adversity.
The short history of global living conditions and why it matters that we know it - by Our World in Data
Max Roser's article challenges the pervasive pessimism about global progress, illustrating significant improvements in global living conditions over the last two centuries. Despite public skepticism, evidence shows substantial reductions in poverty, advances in literacy, health improvements, and political freedom expansion. The piece emphasizes the importance of recognizing and understanding these achievements to inspire continued efforts toward solving the world's remaining challenges, highlighting that progress is possible and ongoing.
Disclaimer: All article summaries are ChatGPT generated. Never rely on summaries - always read the primary source
Book Recommendation
“The Power Law - Venture Capital and the Art of Disruption” - by Sebastian Mallby
About Author: After training as a robotics engineer, I spent nearly a decade as a consultant working on strategy development for global corporations and startups
Now I spend my time strategizing on the Future of Tech, building systems to faster identify and understand new technologies, and thinking about how new technologies will impact sectors and business models across industries
I’m also building a global angel investor syndicate focused on deep tech
You can also find me and more of my thinking on: LinkedIn, Twitter, Website