@levelsio It’s basically because in the US there’s strong regulations through IECC to have a minimum for thermal insulation, hvac, window heat control, etc.
The us made it mandatory for new construction and remodels to have this which is a complete 180 from EU regulation.
@crypto_hades I started a refactor of my original infra to rust to learn and en I estimate around 80k~. Maybe not the best quality since I’m still learning rust but in line with most of what you wrote. Great post
Running DeepSeek R1 on my desk
Uses @exolabs with Thunderbolt 5 interconnect (80Gbps) to run the full (671B, 8-bit) DeepSeek R1 distributed across 2 M3 Ultra 512GB Mac Studios (1TB total Unified Memory).
Runs at 11 tok/sec. Theoretical max is ~20 tok/sec.
AGI at home
Running DeepSeek R1 across my 7 M4 Pro Mac Minis and 1 M4 Max MacBook Pro.
Total unified memory = 496GB.
Uses @exolabs distributed inference with 4-bit quantization.
Next goal is fp8 (requires >700GB)
People mentioned GPT-4o mini is waaay cheaper:
It's 33x cheaper than GPT-4o:
$0.150 / 1M input tokens
$0.600 / 1M output tokens
So that means not $25/mo but:
$0.75/mo for generating 1 million bot replies
$750/mo for generating 1 billion (!) bot replies
Unless we come up with a solution, social media will get exponentially more flooded with bot replies
Tiny Glade is out on Steam! It is written in Rust and they're using Bevy ECS, which is of course exciting for us. Absolutely stunning game. They deserve the success they're seeing!
GIVEAWAY TIME!!! 🗣️🗣️🗣️
Thanks to @IronsidePC we're giving away a prebuilt PC! YOU can enter the giveaway by liking and rting the post as well as following me and ironside ^^
Will pick the winner next week! Good luck everyone :D
This is huge. I’ve been waiting for the Smallville experiment to go open source since the paper came out.
Not only does this have tremendous implications for gaming, but imagine hooking each agent up with a social media account so you could follow along with their daily reflections, behavior, and updates. Imagine assigning different agents different political beliefs and dispositions and observing how they co-exist. Imagine equipping agents not only with the ability to socialize and fall in love, but to reproduce or even to injure or kill one another.
The potential for multi-agent simulations is truly endless, and it’s beautiful to see the basis for future experiments is now open sourced.