These AI advances are particularly interesting to me as they demonstrate models extending beyond traditional text-based applications. ZUNA1.1 is quite neat for bringing EEG processing to variable length recordings. Robostral Navigate shows how vision-language models can steer robots with quite basic hardware. https://stickhook.io
Kimi K3 looks like a credible step toward frontier-quality open weights—especially in coding/agentic areas. The “recursion” angle (coding agents improving training + math/evals improving verification) is where I think the biggest long-term value is. https://wheelielife.io/
I find these developments in AI especially interesting because they show how models are moving beyond traditional text-based applications. ZUNA1.1 expanding EEG processing to variable-length recordings is impressive, while Robostral Navigate shows how vision-language models can guide robots using relatively simple hardware. It reminds me of https://soflowheelielife.org/ a motorcycle wheelie simulation game where adapting to different conditions and responding precisely to what's happening on screen are central to the experience.
"If base-model intelligence becomes commoditized, they can still win by controlling much of the inference capacity needed to deploy it."
They also have to pay for it. And many people are saying they won't be able to.
Here's the bottom line: China is 3-4 times the US population. Their population is educated and hard-working in real productive endeavors - not finance and law. They do more STEM than anyone. They have the biggest real economy in the world, not an AI Bubble economy.
These AI advances are particularly interesting to me as they demonstrate models extending beyond traditional text-based applications. ZUNA1.1 is quite neat for bringing EEG processing to variable length recordings. Robostral Navigate shows how vision-language models can steer robots with quite basic hardware. https://stickhook.io
Kimi K3 looks like a credible step toward frontier-quality open weights—especially in coding/agentic areas. The “recursion” angle (coding agents improving training + math/evals improving verification) is where I think the biggest long-term value is. https://wheelielife.io/
I find these developments in AI especially interesting because they show how models are moving beyond traditional text-based applications. ZUNA1.1 expanding EEG processing to variable-length recordings is impressive, while Robostral Navigate shows how vision-language models can guide robots using relatively simple hardware. It reminds me of https://soflowheelielife.org/ a motorcycle wheelie simulation game where adapting to different conditions and responding precisely to what's happening on screen are central to the experience.
"If base-model intelligence becomes commoditized, they can still win by controlling much of the inference capacity needed to deploy it."
They also have to pay for it. And many people are saying they won't be able to.
Here's the bottom line: China is 3-4 times the US population. Their population is educated and hard-working in real productive endeavors - not finance and law. They do more STEM than anyone. They have the biggest real economy in the world, not an AI Bubble economy.
Do the real math.