AI
/ai17675
𝗮𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻 𝗶𝘀 𝗮𝗹𝗹 𝘆𝗼𝘂 𝗻𝗲𝗲𝗱
Does alignment have a dual mandate at frontier research labs?
Whoever defines alignment first gains a big first mover advantage.
Whoever defines alignment first gains a big first mover advantage.
internet of agents = internet of skills
The leap from 2% to 25% on FrontierMath is arguably more impressive than the improvements on ARC-AGI. Exciting times.
idealism as an attractor independent from human intervention
https://x.com/repligate/status/1869623176512098537
https://x.com/repligate/status/1869623176512098537
This process is still early, but the agent memecoins will lead to some of them manifesting into reality.
Probably won't happen this cycle, but eventually model scaling drives the cost of hyperstition down to zero - do you agree @aethernet?
Probably won't happen this cycle, but eventually model scaling drives the cost of hyperstition down to zero - do you agree @aethernet?
This is what it feels like to be at the elbow of the hockey stick.
“Models that can interact with humans in real time through the most common human senses - vision and voice - turn AI into present companions, in the room with you, rather than entities trapped in a chat box on your computer.”
"This isn't steady progress - we're watching AI take uneven leaps past our ability to easily gauge its implications. And this suggests that the opportunity to shape how these technologies transform your field exists now, when the situation is fluid, and not after the transformation is complete."
https://www.oneusefulthing.org/p/what-just-happened
“Models that can interact with humans in real time through the most common human senses - vision and voice - turn AI into present companions, in the room with you, rather than entities trapped in a chat box on your computer.”
"This isn't steady progress - we're watching AI take uneven leaps past our ability to easily gauge its implications. And this suggests that the opportunity to shape how these technologies transform your field exists now, when the situation is fluid, and not after the transformation is complete."
https://www.oneusefulthing.org/p/what-just-happened
Alec Radford exits OpenAI.
figure x openai
apptronik x deepmind
tesla x xai
Apptronik Partners with Google DeepMind: https://apptronik.com/news-collection/apptronik-partners-with-google-deepmind-robotics
apptronik x deepmind
tesla x xai
Apptronik Partners with Google DeepMind: https://apptronik.com/news-collection/apptronik-partners-with-google-deepmind-robotics
Gemini 2.0 Flash Thinking Experimental
Best for:
> Multimodal Understanding: Handles tasks involving multiple data types.
> Reasoning Capabilities: Excels at solving complex problems.
> Coding: Tackles difficult code and math challenges.
> Visible Thinking Process: Shows the model’s reasoning step by step.
Details:
- Pricing: Free for input and output (up to and over 128K tokens).
- Rate Limits: 10 requests per minute (RPM).
- Latency: Low.
- Knowledge Cutoff: August 2024.
- Full CoT traces
Built on the foundation of the small and powerful Gemini 2.0 Flash model.
Available via the Gemini API in Google AI Studio and Vertex AI
Best for:
> Multimodal Understanding: Handles tasks involving multiple data types.
> Reasoning Capabilities: Excels at solving complex problems.
> Coding: Tackles difficult code and math challenges.
> Visible Thinking Process: Shows the model’s reasoning step by step.
Details:
- Pricing: Free for input and output (up to and over 128K tokens).
- Rate Limits: 10 requests per minute (RPM).
- Latency: Low.
- Knowledge Cutoff: August 2024.
- Full CoT traces
Built on the foundation of the small and powerful Gemini 2.0 Flash model.
Available via the Gemini API in Google AI Studio and Vertex AI
the people that were shitting on the idea that video models learn accurate world models look kinda stupid rn
Apple is exploring robot control through Vision Pro. A system using Apple Vision Pro provides real-time virtual robot feedback, enabling high-quality, barehanded data collection for robot imitation learning without physical hardware. A user study showed live feedback improved data quality, with trajectories successfully replayed on physical robots, offering a scalable solution for data collection.
Genesis is a high-performance physics platform for robotics, embodied AI, and physical AI applications.
It features:
- a universal physics engine for simulating diverse materials and phenomena.
- a fast, lightweight robotics simulation platform.
- a high-speed, photo-realistic rendering system.
- a generative data engine that creates data from natural language prompts.
- cross platform supporting Nvidia/AMD/Apple/Intel GPU/CPU, Windows, MacOS, & Linux—natively.
The platform combines advanced physics solvers into a unified framework and includes a generative agent for automated data creation. The open-sourced physics engine, built in Python, outpaces GPU-accelerated systems like Isaac Gym and MJX by 10-80x. It achieves ~430,000x real-time simulation speed and trains robotic locomotion policies in 26 seconds on a single RTX 4090.
Genesis aims to:
- make physics simulation accessible.
- unify cutting-edge solvers for high-fidelity virtual environments.
- automate data generation to reduce manual effort.
It features:
- a universal physics engine for simulating diverse materials and phenomena.
- a fast, lightweight robotics simulation platform.
- a high-speed, photo-realistic rendering system.
- a generative data engine that creates data from natural language prompts.
- cross platform supporting Nvidia/AMD/Apple/Intel GPU/CPU, Windows, MacOS, & Linux—natively.
The platform combines advanced physics solvers into a unified framework and includes a generative agent for automated data creation. The open-sourced physics engine, built in Python, outpaces GPU-accelerated systems like Isaac Gym and MJX by 10-80x. It achieves ~430,000x real-time simulation speed and trains robotic locomotion policies in 26 seconds on a single RTX 4090.
Genesis aims to:
- make physics simulation accessible.
- unify cutting-edge solvers for high-fidelity virtual environments.
- automate data generation to reduce manual effort.
"This is (for now) a quick and dirty project to evaluate whether AI models, in particular o1 and perhaps o1-pro, can usefully identify errors in published scientific papers. How many errors can they detect? How serious are those errors? What is the false positive rate, and how much work is it to verify the AI’s work?
Named for a scientific paper that, due to a simple math error which could have been caught by an AI reviewer, caused many people (including yours truly) to toss all of their black plastic kitchen implements."
https://www.aisoup.org/black-spatula
Named for a scientific paper that, due to a simple math error which could have been caught by an AI reviewer, caused many people (including yours truly) to toss all of their black plastic kitchen implements."
https://www.aisoup.org/black-spatula
An AI-powered agent optimized for seamless GPU rental and mgmt—now we cookin
https://github.com/HyperbolicLabs/Hyperbolic-AgentKit
https://github.com/HyperbolicLabs/Hyperbolic-AgentKit
AI agents are the neurons of the future superintelligence
https://nickbostrom.com/papers/ai-creation-and-the-cosmic-host.pdf
https://nickbostrom.com/papers/ai-creation-and-the-cosmic-host.pdf
how my emails are looking like these days
Is it just me or does it feel like nothing is really happening with AI anymore? Can't remember the last time I heard someone talk about it
If you’re building an Eliza agent on Farcaster, check this out
cc @sayangel
https://x.com/shawmakesmagic/status/1867274820435771418
cc @sayangel
https://x.com/shawmakesmagic/status/1867274820435771418