You still have time to abandon all belief in intelligence.

Ok small rant: I don't think we're in a generative AI bubble. Actual market is still small. It seems bigger only because all kinds of things (and liabilities) are labelled "AI" now. The long overdue infrastructure update? For AI. Having to cut jobs as revenue outlook is not

Grok 2 from xAI has just been released on Hugging Face: huggingface.co/xai-org/grok-2

If you're interested in radars, there's an article in the Sep 25 issue of IEEE Microwave Magazine, on design of integrated (CMOS) phased-array FMCW radar transceivers. It covers some basics, and then goes through a few designs from 5.8GHZ to 94GHZ. DOI: 10.1109/MMM.2025.3539245

DSPy-ians may disagree, but I don’t think automatic prompt optimization is the real killer feature of DSPy. For me, the most important value proposition is the speed of experimentation it enables and I’m convinced that this will further increase (as dspy gets even better).

i have to set up batch ray tracing rendering on cuda infrastructure on my server configuration this weekend wish me luck

it's basically over. they've won. full vertical integration. TPUs. google level infrastructure. everything just works. work life balance. all the good researchers i know end up there given enough time

the zoomer goes to openai and makes 3 million dollars in a year and then retires and hacks on cool shit while travelling through europe pipeline

this is actually new, it never did before

maybe it did and i never noticed and now i'm noticing because i'm taking creatine again

got a mysterious order from china that said "wires, cables" on the label until i opened it up to find the computer chips i ordered. must have just been an honest mistake!

imagine buying a car that you have to have a subscription to use when i found this out i lost my mind

j-just one more payment bro. 200$ a month for my car that's already paid off. or else the GPU inside it that i paid for becomes a brick of weight

how the esp32-s3 looks at me when i say i want to build in america

love all of you. we're all gonna make it, good night

Gemma 3 270M fine-tune on Windows with Unsloth AI. It worked! 🥳

if web development is mostly centering divs, and theoretical CS is just Chernoff bounds, then it seems like ML systems is mostly just AllReduce + lots of coffee

This is the number of param of grok 2 according to grok4 - Total params: 285.62B (tied) or 286.70B (untied) - Active per token: 78.39B (top-2 experts) - Passive per token: 206.16B (the unused experts)

every engineer should be doing endurance sports

on demand regime change powered by web3 and agentic ai