AI News & Talk & Personal builds for Home

A laser in a lab in eastern China is now writing full color images directly onto bare metal, ten square centimeters every second, and not one drop of ink ever touches the surface. No cartridge. No nozzle. No pigment at all. Just a beam of light and a sheet of metal that comes out...See more

AirLLM makes it possible to run 70B-parameter AI models on consumer hardware by loading models layer by layer from storage instead of all at once. Combined with FlashAttention, it lowers the barrier for students, researchers, and indie developers to use models like Llama 3.3 70B on a MacBook or gaming PC. #AirLLM #FlashAttention #Llama33 #OpenSourceAI #LocalLLM https://www.instagram.com/reel/DZEKM7iOXQw/?igsi=NTc4MTIwNjQ2YQ==

That cracked phone in your drawer outspecs an $80 Raspberry Pi. Hardware Haven took a worse-for-wear Xiaomi Mi A1, pulled the swollen battery, fed 5 volts through a diode into the charge controller, installed postmarketOS and turned it into a home server running Home Assistant -...See more

AI Just Learned How to Make Real Phone Calls AI agents can now make real phone calls instead of just sending messages. Using xAI Voice Agents, you can build your own AI caller by simply describing what you want it to do. Choose what the agent should say, pick a voice, get a...See more

The Biggest AI Model Just Ran on a Laptop Nvidia’s CEO might want to look away. Building You 100X. Someone just found a way to run a 2.78 trillion parameter AI model without a GPU. One CPU. Just 8GB of RAM. Zero graphics cards. And the entire engine powering it is...See more

Datacenters are quietly selling their old AI cards for scrap prices. The NVIDIA Tesla P40 launched at $5,699 as a datacenter AI card. Today pulls go for $150-200 on eBay. Four of them make 96 GB of VRAM, enough to run 70B-class language models at home. The documented build: a...See more