Pinned
Announcing Flapping Airplanes!
We’ve raised $180M from GV, Sequoia, and Index to assemble a new guard in AI: one that imagines a world where models can think at human level without ingesting half the internet.
GIF
Flapping Airplanes
59 posts
Flapping Airplanes
@flappyairplanes
AI lab
- Replying to @flappyairplanes(5/5) We think it’s not the end of history when it comes to new algorithms, and history suggests important new algorithms come alongside new systems. We paid for the whole GPU; we’re going to use it!
- Replying to @flappyairplanes(4/5) One thing we’ve built is a “kittens” virtual machine that takes over the whole GPU and allows new kinds of co-optimization. We can go past the traditional sequential kernel model – for example, fusing entire training runs into a single kernel and even weirder stuff.
- Replying to @flappyairplanes(3/5) We want algorithms that are >1,000x more data efficient. These algorithms look like they’re going to be weird and different enough to require new systems under the hood. We’re still not ready to talk about these algorithms, but we’re excited to talk about our systems!
- Replying to @flappyairplanes(2/5) LLMs are great at high data tasks like search and coding but struggle with low data tasks including trading, science, robotics, and more. FLOPs are getting cheaper over time, but new data is becoming more expensive to collect.
- (1/5) Great to be at @sequoia to give a sneak peek of one of our research directions! TL;DR one path to data-efficiency may be to “abuse GPUs like they’ve never been abused before”
00:00 - Flapping Airplanes reposted
Flapping Airplanes co-founder @amspector100 explains why data efficiency is the greatest bottleneck to AI adoption: "To the extent that AI has been hard to integrate into the economy, I really think it's because models are much less data-efficient than humans. If you want it to
00:00Replying to @tbpnvery proud to have come onto the show to defend our two core principles: data efficiency and whole milk - Replying to @flappyairplanes
- We are also truly honored to have @SuhailNimji @jml_campbell as advisors. They are true company builders who know the valley inside and out but more importantly they know how to help young people believe in themselvesThis isn’t just another Neo lab. @bfspector & @amspector100 are forces of nature. A dynamic duo defined by grit, drive, and authenticity—two of the most charismatic, thoughtful, and genuine individuals I’ve met. Honored to be an advisor - @flappyairplanes
- let's give the models a childhood ✈️Current AI needs the whole internet to learn; humans just need a childhood. I’m joining Flapping Airplanes to help close that gap. We’re building models that think deeper by reading less. It’s time to fly over the data wall—excited to show you what’s next! ✈️
- Flapping Airplanes reposted
I spent a bunch of time a year ago thinking about the data wall. A blackpill at the time for me was when I realized that the total stock of natural text data is depleting much faster than Chinchilla's infamous 20 tokens per param compute optimal ratio suggested. Here is a naive
A conventional narrative you might come across is that AI is too far along for a new, research-focused startup to outcompete and outexecute the incumbents of AI. This is exactly the sentiment I listened to often when OpenAI started ("how could the few of you possibly compete with - Flapping Airplanes reposted
Announcing Flapping Airplanes! We’ve raised $180M from GV, Sequoia, and Index to assemble a new guard in AI: one that imagines a world where models can think at human level without ingesting half the internet.GIF
