Sebastian Hindhede (@MrSuperSecret) header
Sebastian Hindhede (@MrSuperSecret) profile picture

Highly passionate about applying high performance, highly scalable and secure systems to solve real problems and innovate. CTO, senior developer, adventurer.

Danmark
Joined September 2015
Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)Photo by Sebastian Hindhede (@MrSuperSecret)
Sebastian Hindhede 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.
269
569
326
4,587
478,914
Sebastian Hindhede retweeted
Introducing iOS for the web. This is Telegram (written in Swift and Objective-C), running fully in a browser on our custom renderer. No emulator, No iPhone required. You can infinitely branch the entire codebase by copy pasting a frame and make changes to each frame. We built a compiler+renderer (in Rust) that allows you to easily design on top of a native iOS codebase. Most edits land in under a second and is much much faster then xcode builds. Our goal is to take any codebase and turn it into a design environment that fully runs in your browser!
281
318
125
5,243
398,450
Sebastian Hindhede retweeted
RustFS now supports RDMA — with S3 over RDMA. Final testing is in progress. If all goes well, we'll release it and open-source the work.
🤖 Made with AI
17
4
172
13,944
Sebastian Hindhede retweeted
We just settled the DOM-based vs WebGPU canvas debate. This is our WebGPU canvas rendering ~7,000 real HTML/CSS frames with over 2 million elements, at 120 fps. And it's not just static designs. Every frame runs your full app so you can click around etc. We built our own HTML renderer that supports every single CSS property (blurs, shadows, gradients, 3D transforms, you name it). We have full parity with chrome's own renderer! This is possible because we own the entire pipeline. Our custom bundler takes almost any codebase and splits out its HTML, CSS and JS that is optimized for our canvas. We basically built a mini browser, in Rust, running inside your browser. Two years of infra work compounding and coming together finally 🥲
165
188
34
3,173
158,314
Catala is seriously underrated. Imagine formalizing the entire tax law of a country and compiling it down to executable code - because that's exactly what you can do with it. Companies spend millions upon millions to convert tax law into code, over-and-over, they absorb the maintenance tax. Imagine if we just erased all of that and instead distributed Law as a Library.
30
Sebastian Hindhede retweeted
Tiered storage achieved internally. Releasing next week. Infinite data per database.
4
3
49
1,673
Ohh boy
Good morning, we escaped the famous Firecracker VM and got root execution on the host. Full firecracker escape technical writeup and Proof of Concept code attached below.
2
50
Sebastian Hindhede retweeted
We are really excited to present our full paper, LinkHydro, at #UIST2026! LinkHydro enables rapid fabrication of soft self-pumping tubes in approximately 10 minutes using heat-shrink tubing for prototyping! Full video: youtube.com/watch?v=g-XVmdD9…
17
62
9
624
33,168
Sebastian Hindhede retweeted
1/4 New writeup: you can fingerprint the exact GPU you're running on with a default torch.index_add_ call, even among identical GPUs. This works because floating-point maths depends on order, and every GPU has its own. This fingerprint survives reboots and can't be faked.
14
88
19
913
84,456
Vertical agents is the perfect business opportunity right now - agentic tech is extremely capable, easy to build with if you know how and there are plenty of local businesses that could benefit from them, but don’t even know who Claude is
1
1
101
This is the way
Replying to @hichaelmart
a really good pattern I started using is "digital twins" recreating the API surface of third party services you depend on, making them run locally on top of sqlite for example I made a digital twin of the Discord API to test my kimaki bot github.com/remorses/kimaki/t…
1
81
Agents need digital twins, mark my words. Clone everything into digital twins and I mean literally everything. Especially useful for simulation testing!
31
AI is learning to mathematically prove the software it builds. That’s bigger than faster coding. Forget dark factories. I want LIGHT factories. Every bit visible. Every claim verified.
Making sure I state it clearly: We can literally patch and prove bug-freeness of the entire software stack running in production. With AI that exists right now for ~$100m. This means no more software-level hacks. I know it sounds too good to be true, but it isn't. AMA.
38
Where do your agents reside? On your laptop? On your phone? At X, Y, Z provider? Locally? In a sandbox? If you DIDN’T answer in a database/in a bucket, you have fundamentally confused execution environment with memory. All an agent is, IS its sum of its memories. If you are prompting it in a chat on your phone, it should “live there”. Or your smartglasses. Or your Mac or Omarchy laptop. Or your cloud/sandbox. If you are building agents and this is not clear as day to you, you run into stupid scenarios like: - Executing untrusted code on whatever device you are prompting on. - Wasting cloud compute on inference heavy sessions. - Splitting valuable context between machines. In other words, you accidentally treat the place your agent works with the place it resides. Your agent resides where its state, where its memories and where its context is stored. Decouple that from its work environments. This is the way to truly build ambient intelligence, which is what we all truly want. Ambient intelligence is when your agent or teams of agents seamlessly move with you, spinning up in the environment where you meet them (or go back to their mothership when you are done interrupting them, to do important agent shit, while you watch baseball or whatever). The agent seamlessly moves to where it does its work best. Its memories do not.
1
44
This is extremely fascinating. What if information is a field or a dimensional space we tap into and by training models, we imprint that same field into vectors. The training data we feed these models may simply be the downstream lossy decoding of that supposed information field. Or perhaps reality is intrinsically mathematical in its very structure.
This quoted post is unavailable.
1
1
74
There is also the slightly boring version: at the end of the day, labs train on more or less the same data, produced by more or less identical observers of the same physical and subjective reality. Perhaps convergence toward similar mathematical structures is simply the inevitable result.
1
3
34
In other words: The models may not be converging on the structure of reality itself. They may be converging on the structure of humanity's collective representation of reality.
4