Crazy stat from the Llama paper: > For Llama 3 405B , we noted a diurnal 1-2% throughput variation based on time-of-day... the result of higher mid-day temperatures impacting GPU dynamic voltage and frequency scaling. 2025 jobs be like "Applied Metereologist, Pretraining"

I retrained the gesture recognition model last night. You can now draw with a more natural writing pose (like holding a pen) making it a lot easier to interact with the whiteboard! Switching tools is also a lot more responsive. Try it here: floating-hands-whiteboard.vercel.app

Did somethin today. 💯👊🏿🔥 #FollowTheDogg

From primordial soup to life in a virtual environment. One of the most exciting papers I have read in long time studies how artificial self-replicators emerge spontaneously in a virtual artificial chemistry: arxiv.org/pdf/2406.19108 by Blaise Aguera, Jyrki Alakuijala,

Dopamine fasts are growing on me. Five minutes away from your phone or computer: time stretches and boredom kicks in. When you return, work feels exciting

Very proud of this work, great to see it published! Giving rating systems (think Elo) the full Bayesian/state-space model treatment 🥳

We're hiring in Copenhagen! I'm starting a new office for Normal Computing in Denmark in October and hiring a founding Staff AI Engineer to work on NLP, LLM Reasoning, and Verification of chips directly from the specification. As a Staff AI Engineer, you will help build a

hey anon do you want to build chips with your friend zach? apply here: jobs.ashbyhq.com/Normal%20Compu…

Our team is hiring hardware engineers in London! Come join us if you want to build unconventional hardware and have a big impact on the next wave of compute

Youtube is getting cool again youtube.com/watch?v=AWtC4n…

As Moore's law runs out of steam and machine learning heats up, the race is on to find new ways to compute—and thermodynamic computing is a hot new contender. Unlike traditional microprocessors, which rely on binary logic and electrical circuits to process information,

Complex step approximation is a numerical method to approximate the derivative from a single function evaluation using complex arithmetic. It is some kind of “poor man” automatic differentiation. nhigham.com/2020/10/06/wha…

OK, here's a very simple calculation showing that adding a bit of randomization can make numerical integration better even in the one-dimensional setting.

This is a beautiful plot. Any more of these existing anywhere?

Join Sam Duffield and myself today at 1 pm UK time for a seminar on “Thermodynamic Computation and Applications in Machine Learning” at the Alan Turing institute to learn about the cool stuff we’re working on at Normal Computing 🧠🌡️ 😎 Zoom link: turing-uk.zoom.us/j/99676418386?…