🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Since covid, I’ve been thinking about what the next black swan event could be. One possibility is a solar storm stronger than the Carrington Event of 1859. It would likely disrupt the electric grid and satellite comms, but it might also open a window for a bad actor to take

I recently learned that nutmeg used to be worth more than gold. In the 1600s, the Dutch traded Manhattan to the British for a tiny Indonesian island that grew it.

delightful rabbit hole: you may know what a luthier is (violin maker), but likely haven't heard of an archetier (bow maker). archetiers play a critical-yet-overlooked role in the art of violin making. They use a rare & endangered brazilian hardwood called pernambuco to make

in 10 years, I think relationship-mediated investments will be mediated by AI (10x faster and more rational)

🚀 Supervised Fine-Tuning (SFT) has been the default for adapting LLMs—but it has a major limitation: it demands a large amount of high-quality labeled data. 🧵 Why Reinforcement Fine-Tuning (RFT) is a better approach when data is scarce: 👇

So excited to have our partner Sundeep Peechu lead this big round for Mercor partnering with the awesome founders Brendan (can/do) adarsh surya for such an ambitious vision to have AI assess talent and maximize its potential… cc James Detweiler techcrunch.com/2025/02/20/mer…

Introducing Carl, the first AI system to create a research paper that passes peer review. Carl's work was just accepted at an @ICLR_conf workshop on the Tiny Papers track. Carl forms new research hypotheses, tests them & writes up results. Learn more: autoscience.ai/blog/meet-carl…

Today we're thrilled to announce the first end-to-end platform for Reinforcement Fine-Tuning. With just a dozen labeled data points, you can outperform #OpenAI o1 and #DeepSeekR1 on complex tasks. Built on the #GRPO methodology that DeepSeek-R1 popularized, our platform delivers

Robots won’t be truly helpful until they can navigate like humans. Tera AI is solving this with software-only spatial reasoning—helping robots move using just a camera, no extra sensors needed. James Detweiler and I are excited to back Tony Zhang and @TeraAI_ at @Felicis!

This is my contrarian hot take of the moment. (Which is apparently not as contrarian as I thought.)

Inspired by the success of LLMs, today on the blog we discuss how neural activity in the human brain aligns linearly with the internal contextual embeddings of speech and language within LLMs as they process everyday conversations. Learn more →goo.gle/4iiUoNj

Congrats to Felicis' James Detweiler, who was named one of Venture Capital's Rising Leaders by The Information. James invests in the boldest AI companies — Mercor, Skild AI, Flower, & more — that aren't afraid to challenge the status quo or invent new markets entirely.

Big kudos and well deserved for James Detweiler ! Featured as one of Venture Capital Rising Leaders by The Information

We’re excited to share James Detweiler has been promoted to Partner at @Felicis! James shows up every day with conviction, optimism, and energy - just ask @brendanfoody, Deepak Pathak, vik, nic lane. If you’re tackling a “root node” problem in AI, like robotics infra,

Just landed, ready for ICLR. Below is a run down of the work and events from the Cambridge ML Systems Lab (CaMLSys) and Flower Labs you can see this year. 🔬DEPT: Decoupled Embeddings for Pre-training Language Models = Oral Session 5B: Sat 26 Apr 11:30 a.m = Poster: Hall 3 +

A long time ago, in a cloud far, far away… Devs begged to run VLMs locally. Now the rebellion begins. 🛰️ Moondream Station — run Moondream on your Mac, no cloud required. CLI or local port. No-hassle setup. Totally free. moondream.ai/station May the prompt be with you.

when you realize your 2b vlm has better benchmark scores than llama 3.2 90b

I applaude everyone who's investing their energy into building the future We need more optimism, more people building, more progress Less arguing about current affairs, less fixed pie thinking, less doomerism The future belongs to those who build it

The U.S. spends $4.9T on healthcare, more than any country, and gets worse outcomes. Prior authorization is a big part of the problem. It’s painful, complex, & full of error-prone tasks. This can be fixed by AI. More from James Detweiler & Eric: felicis.com/insight/health…

We believe the future of AI will be decentralized. Join us for Flower AI Day in San Francisco to learn what’s next.