Jonas Metzger
349 posts
@JonasMtzgr
I try to compute economic problems away. Quant Researcher @ Citadel. Prev: PhD Econ, CS @Stanford.
Wired is reporting that OpenAI is preparing to launch a stand-alone social media app for Sora 2. The app is a vertical video feed with swipe-to-scroll navigation, just like TikTok, except the content of this app is 100% AI-generated.
I’m stoked to share our new paper: “Harnessing the Universal Geometry of Embeddings” with @jxmnop, Collin Zhang, and @shmatikov. We present the first method to translate text embeddings across different spaces without any paired data or encoders. Here's why we're excited: 🧵👇🏾
Security researchers have uncovered a pre-installed, undocumented remote access tunnel in 🇨🇳 Unitree Go1 robot dogs. Each Unitree Go1 robot dog is shipped with a preconfigured tunnel client that initiates a connection to 🇨🇳 CloudSail — a remote access platform developed by 🇨🇳 Zhexi Technology, based in China. “Anybody with access to the API key can freely access all robot dogs on the tunnel network, remotely control them, use the vision cameras to see through their eyes, or even hop on the RPI via SSH.” “Most of the machines are located in China, but as expected some are outside of China, apart from some residential IPs, we were able to identify several University IPs and some corporate networks from around the world.” More than a dozen universities from the US, Canada, Germany, New Zealand, Australia, and Japan have experimented with Unitree Go1 robot dogs: USA: MIT, Princeton University, University of Massachusetts Amherst, Carnegie Mellon University Canada: University of Waterloo Germany: Hochschule Coburg New Zealand: University of Otago Australia: UNSW Sydney, Deakin University Japan: Shinshu University The discovery raises serious concerns about supply chain trust, especially as these robots are widely used in academic, corporate, and even defense-related environments. cyberinsider.com/remote-access-…
If the light blue part is best of N scores, this means that Grok 3 reasoning is inherently an ~o1 level model. This means the capabilities gap between OpenAI and xAI is ~9 months. Also what is the difference between "think" and "big brain"
We’ve found as AIs get smarter, they develop their own coherent value systems. For example they value lives in Pakistan > India > China > US These are not just random biases, but internally consistent values that shape their behavior, with many implications for AI alignment. 🧵
