alvin

anyone at yc??? wld love to meet you!

yhhh okayy we’ll be at @ycombinator soonn

which can be loaded into cache somehow but even if we managed to do that we could only do that for one token after activation vector is multiplied with weights its useless and should be disregarded because its in cache and you need it again when you do the same ops another token

Modern Computer architecture works on assumption that if you access X, then you'll prob access X again sooner and it works like magic fuck that was genius but when itcomes to llm for the the weights size is in megabytes

Ran a 1B parameter LLM on my CPU and profiled it. 30 seconds total. 0.058 seconds of actual compute. 39.76% cache miss rate. The CPU spent 99.8% of the time waiting for data. Inference isn't slow because your CPU is weak. It's slow because weights can't move fast enough from RAM.

wrote forward propagation by hand on paper today just to actually understand it. not gonna lie derivatives and the idea of slope finally clicked. building toward making inference cheaper on my RX 7600. Maybe JUST maybe its a pipedream but whatever...

@portal2urmom @grok whats ghe context

Everyone! Free up like, 61.67GB on your drive! Now!

probably the cleanest context engineering guide that i've read in a while... github.com/humanlayer/adv…

Nassim Taleb, damn

been so longg

i wanna redesign the whole internet

kinda got employed latelyyy

@Hi_Mrinal we all know that you’ll get it 🤌

Applied to Anthropic as a SDE

@WildCat_io @mazeincoding @grok wtf does this mean

@mazeincoding capcut: - expensive - stealing data / copyright - slow and buggy opencut: - free - opensource - founder is a great ** poster on X!

claude: - expensive - doesn't follow instructions - is lazy gpt 5: - cheaper - follows instructions - is not lazy GPT 5 IS BETTER THAN CLAUDE

@RubenVeidt lately i’ve been loving your posts a bit too muchh

to appreciate css, one must write vulkan , to use vulkan to the max, one must understand css this 2 line of css .container { height: 100vh; } to do the same in vulkan, you need to Initialize the vulkan library, select a physical gpu, create a logical device and command queues, create a window surface and a swapchain to present images to it, setup up a render pass that describes the structure of a rendering operation, define the entire graphics pipeline, including vertex shaders, fragment shaders, rasterization state, blending state, etc...., create framebuffers, allocate and manage memory for the vertex data, create command buffers and record the exact drawing commands, use fences and semaphores to synchronize the cpu and gpu to make things happen in the correct order this is not even recreating the css stuff, its just setting up vulkan to even start doing that the more you understand what's happening under the hood, the more you appreciate when and what you don't have to think about it

I seee…

broo cookeed something cooool

@theo its another way of procrastination lol

I really hate the “market your product before building it” trend.