Compound (@CompoundVC) header
Compound (@CompoundVC) profile picture

Compound is a thesis-driven, research-centric investment firm | What we're currently reading @compoundarxiv | Not called Compound VC

NYC
Joined June 2009
Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)Photo by Compound (@CompoundVC)
Compound retweeted
Introducing AIR-O, a model that decodes what a person is attending to directly from brain scans. Using only noninvasive brain data, we can generate an image from scratch in real time that depicts where a person is focusing their attention. This is the first of a new generation of algorithms that can read the shape of thoughts directly from brain activity.
242
490
263
4,093
408,485
Team @CompoundVC is building a really special firm - I feel grateful to be a small personal investor and love when we get to co-invest in opportunities together Congrats on your first 10 years - excited to see the types of ambitious and unique founders and opportunities you’ll back over the next 10 @mhdempsey @shelbynewsad @0xsmac @mackenziejem
8
1
90
5,184
Compound retweeted
Today, we’re making Praxis-1 available to everyone. Our state-of-the-art world action model turns video pretraining into control for real robots, and it ships with open weights under Apache 2.0, ready to run and fine-tune on your own hardware.
129
97
126
622
103,128
Praxis-1, our first world action model, is now available for anyone to use. It is currently the leading open-weights robotics model on the RoboLab leaderboard and outperforms every other world action model on the benchmark. More than that, it’s versatile and easy to fine-tune for new hardware and tasks. We’ve tested it on 9 different robot embodiments so far, in simulation and on real hardware with multiple partners, and we’ve seen it reach good performance on a new embodiment with under 30 minutes of demonstrations. Weights, code and fine-tuning recipes are on GitHub and Hugging Face. It can run on a single RTX 5090 with NVFP4. We started our robotics work about a year ago with the thesis that the same video pretraining we do for our creative models can be massively useful for building systems that interact with the physical world, and the benefits of transfer are now becoming clear. When I first talked to people in the field, there was a lot of skepticism about whether video models would be useful in robotics at all, and then about whether you could use them as policies. From first principles, it makes sense that if you can predict the future well in pixels, extracting an action representation should be the easy part. Reality caught up with first principles. One of my favorite things in our technical report is the idea of self-rollouts. Because the model’s performance is bottlenecked by its visual imagination, you can quickly get an intuition for whether it will perform a new task well by rolling out the policy in the model’s imagination. That can speed up evaluation, and it opens a path to doing reinforcement learning in the model’s imagination. There’s been a lot of discussion recently about frontier LLMs being the bitter-pilled way to solve robotics. But scaling performance from unsupervised video is an even better example of the Bitter Lesson than scaling from text. Praxis-1 is more than 10x faster and 50x cheaper than GPT-6 Astra on the manipulation tasks we tested. Frontier LLMs will keep improving rapidly on the Pareto frontier of cost and performance, but so will world models. There are also great ways of combining the strengths of the two. And we are much farther from saturating multimodal data than we are text. Thanks to the mighty Runway Robotics team for making this happen, and to our partners for being so generous with their time helping us test the model across a wide range of hardware and tasks.
Today, we’re making Praxis-1 available to everyone. Our state-of-the-art world action model turns video pretraining into control for real robots, and it ships with open weights under Apache 2.0, ready to run and fine-tune on your own hardware.
8
17
3
95
6,231
New Blog Post: The Era Of Fully Autonomous Companies. Fully autonomous companies are now possible. By this time next year we will see millions of them, and many will be profitable. In early December, we'll show you some. generalintelligencecompany.c…
15
14
6
264
18,805
We built Valius for one reason: help patients outsmart cancer. A few months ago, we were privileged to work with the Thornhill family and their remarkable 11-year-old son, Fox, who has osteosarcoma. Together, we’re flipping the script on Fox’s disease.
6
20
11
141
16,607
Compound retweeted
Hear more from CEO @AdamTSilver on the news 👇
1
1
1
15
1,850
Compound retweeted
Alaska and Plural just launched InvestNorth, a new platform for financing infrastructure in Alaska. Bloomberg (@business) has the exclusive: bloomberg.com/news/articles/… (Third-party reporting; statements and figures are Bloomberg’s own.) Securities offered through Plural Brokerage LLC, member FINRA/SIPC.
6
5
5
30
8,532
Compound retweeted
alaska has a rich history of sharing its natural resource wealth with residents the new state-sponsored financing powered by @PluralEnergy means they can do the same for the infrastructure build
Alaska and Plural just launched InvestNorth, a new platform for financing infrastructure in Alaska. Bloomberg (@business) has the exclusive: bloomberg.com/news/articles/… (Third-party reporting; statements and figures are Bloomberg’s own.) Securities offered through Plural Brokerage LLC, member FINRA/SIPC.
4
3
37
3,318
Compound retweeted
NEW POD - Is It Over With @DualityResearch we discuss: > his background > the research process > how he's thinking about the current state of the market > what investor misconceptions exist + a bunch more trad market things links below note: this was recorded october 27
8
4
2
74
6,190
Our engineers start each cold plate design for our Two-Phase Direct-to-Chip Cooling System from a first draft proposed by CurieOS, then look for its weak points. In a two-phase direct-to-chip loop, liquid is pumped to the cold plate, boils in a controlled way and returns to the coolant distribution unit (CDU) as a mix of liquid and vapor, so the fluid's thermophysical properties set the line sizes and the head losses through every fitting. CurieOS, our AI operating system for science and engineering, takes those fluid properties and the system requirements and quickly proposes system specifications and an initial design, which frees our prototype and hardware engineers to put their judgment into choosing the right path. As the hardware goes through each round of iteration, CurieOS works through the changes alongside the team, so revised designs come back much faster. Watch Graham Wenz, our Director of Engineering, explain how it works, then follow the link below to learn more about CurieOS.
4
4
7
535
Compound retweeted
EXCLUSIVE: AI agents are getting their own app store. Cowboy launches from stealth today with $5M, and 100 agents in its marketplace. I spoke to CEO Chad DePue (@chaddepue) about his vision for a safe way to deploy and trade agents that doesn't depend on Anthropic or OpenAI.
15
14
8
73
15,369
RT @mhdempsey: An excerpt i wrote from an LP letter for our public fund: On Misery Each quarter I try to spend a few days reading through…
This quoted post is unavailable.
2
47
Replying to @CompoundVC
@CompoundVC's monthly letter is out including our latest research on clinical trials and quantum And our top reads including a founder detailing how the robotics data sales business is broken + how Periodic integrates its wet lab Sign up below! compound.vc/newsletter
3
13
1,090
New Post: What Is Quantum Advantage Actually Worth? We’re nearing useful utility-scale quantum computers. And yet almost no expert can point to what they’ll be useful for Beyond breaking cryptography, ppl vaguely refer to quantim sim We hope to see more clarity from the field
2
3
6
793
Tragedy formed our modern clinical trial system. Opportunity has the potential to form the future of our clinical trial system. With this piece, we @CompoundVC walk you through the case of thalidomide that was the lightning rod for adverse event reporting, investigation new drug applications, and efficacy trials before drug approval. We also discuss impactful regulatory updates coming to clinical trials in the US which could lead to faster feedback loops. In our field's ambition to eventually cure all disease, we must learn from the past to make a better future.
2
6
2
38
2,959
Compound retweeted
Pedigree is a measure of who got backed in the past. How do we bet on the future instead? This is episode 6 of Confessions on Modern Philanthropy with Michael Dempsey (@mhdempsey). Michael helped build the data and research team at CB Insights, invested in public and private markets, and now runs Compound, a research-centric firm that writes some of the first institutional checks in frontier technology. We asked him what led him to start funding people through small, no-equity grants. Michael on pedigree: "We've done such a good job of backing people with pedigree being able to accumulate capital that actually you should intentionally rotate away from pedigree in certain times." Some highlights: 05:12 Professional and creative soulmates: Matisse & Picasso 07:19 Belief as a form of patronage: being seen before you can see yourself 11:11 What tech underrates about lack of constraint 15:37 Why micro-grants? 19:50 The one question that filters 90% of grant applicants 24:47 The Thiel Fellowship, Goodhart's Law, and getting into YC and not going 25:55 Sacrifice as the signal 31:05 East Coast versus West Coast framings of giving 34:43 Effective altruism lacks romanticism Full show notes: analogue.press/p/confessions…
5
2
17
5,250
Compound retweeted
Introducing Runway App Studio. Build the creative app you wish existed, just by describing it. It generates images, video and audio with Runway’s models and your Brand Kits, ready for you or your whole team. No code required. See examples in the thread below.
67
61
77
369
204,928
A data center submittal package normally two to three weeks to produce. With CurieOS, it can take two to three days, with every calculation recorded so it can be checked by our engineers. Watch Jaden Weaver walk through how it works in the video, then follow the link below to learn more about our AI operating system.
1
2
1
3
402