Today, I’m excited to introduce @arenaphysica. For the past few years, we’ve been quietly partnering with companies pushing the frontier of hardware like @AMD , @anduriltech and @SiversSemicond – deep in the guts of their most complex machines. Where applied physics, specifically the laws of electromagnetism, dictate performance. Electromagnetism is a domain poorly suited to LLMs, and a domain I spent most of my physics PhD trying to understand.
At Arena Physica, we are in pursuit of electromagnetic superintelligence.
We believe that a new class of foundation model will let humans push farther into our understanding of physics and will let us wield forces like EM that shape our world, but are fundamentally unintuitive to humans.
It was an honor to partner with my favorite essayist, @packyM to explain how electromagnetism secretly runs the world
The future is electromagnetic.
One challenge is that there are ~ten people in the world who can deeply intuit electromagnetism. RF engineering is "black magic."
Arena Physica thinks machines can intuit EM better.
CEO Pratap Ranade & I on AI for EM:
https://t.co/U3CUnqcYPp
In November 2022, @markiewagner wrote Choose Good Quests, then she went dark to work on her own.
Today, she's launching Poetic, a new class of software that's adaptive like AI, reliable like code.
This is her first public essay since CGQ, on why & how.
https://t.co/bhpxhiKjVY
Truly inspiring work from @Vbob202, and the entire team @Saronic 👏 – the first rescue of a downed aircrew by an unmanned system.
If this unmanned Corsair were a person, it would have won a medal for bravery
A Saronic Corsair unmanned surface vessel rescued the Army AH-64 Apache crew after they crashed near the Strait of Hormuz in the waters of Oman yesterday, the first-ever rescue of downed aircrew by a drone boat -WSJ
I actually grimaced at this framing. "Here is something super scary, but dont worry we made a cutesy safe one for you - it's named after fantasy land and you love that kind of stuff!"
I really love claude and generally I like the anthropic marketing ethos, but *often* their comms makes me feel talked down to
Distributed batteries are quietly becoming the most important new architecture in American energy. Learn why in Episode 2 of The New Feynman Lectures with Justin Lopas of Base Power Company - live on YouTube, Spotify, and Apple Podcasts!
Highlights:
0:12 - Why batteries move energy through time and wires move it through space.
5:28 - ERCOT as the air traffic control and stock exchange of Texas electricity.
10:00 - What happens to prices when a South Texas gas plant trips offline.
14:54 - Why the grid averages 50% utilization but hits over 100% at peak.
20:58 - Centralized batteries vs. BASE's distributed model on tens of thousands of homes.
22:35 - Virtual transmission: replacing transmission lines with distributed batteries.
@camillericketts@arenaphysica The chips make perfect paperweights. We would have loved to ship it with the devices that bring it to life (power supply and network analyzer) but that would have blown the merch budget
Simulation is the dark horse. Physics sims especially for fields that our biology didn’t evolve to sense will be the biggest and least obvious drivers of enormous market value.
Great decoder ring for all that falls under the world models banner by the legendary @drfeifei
This is a really cool project that @PratapRanade and team are working on: The New Feynman Lectures.
I had the opportunity to share what I've learned about the grid and how our batteries operate within it. 👇
@basepowerco is one of the most interesting startups out there. Base co-founder @JLopas knows what he’s talking about when it comes to batteries – before founding Base he was an engineer at @anduriltech and @SpaceX . I’m incredibly excited to drop our second episode of the New Feynman Lectures by @arenaphysica today with Justin. If you are curious about how the grid works, and how a distributed network of batteries can bring balance to it, then these will be 20 minutes well spent. Link in comments.
@semiDL Interesting, per unit distance though, right? Still doesn't distance dominate, so placing it on die, or on interposer shortens the distance more than compensating for latency/unit length
Latency, not compute is the limiter for next generation AI chips and data centers.
Laying down dense patches of “wires” between the chips for fast I/O is a hard EM problem — electrical signals from nearby wires affect others (crosstalk), causing noise, and at high frequency signals can reflect back (like light off a mirror). The fastest way is to put all of it on a giant piece of silicon (@cerebras), the next best option multiple dies talking to each other across a silicon “interposer”, and the worst is on a copper wire along the green PCB motherboard.
Depending on which path you take, the energy it takes to move one bit can be 70x more.
Only @nvidia and @Broadcom have really mastered this electromagnetic art.
Well explained @NatalieFratto
For the last 70 years, we've been dicing up silicon wafers into hundreds of tiny chips. We had to because it was the only way to cut out the defects inside the silicon.
@cerebras, which IPOed last week (the biggest tech IPO this year so far in terms of capital raised), found an innovative way to use almost the whole wafer and make one BIG chip instead. They built redundancies into the wafer with duplicate parts so that defects didn't prevent it from working.
The latest @chartsncrafts dives into the tech behind it and the zig-zagging path they took to IPO.
PS I really only did this one because I wanted to try my hand at drawing holographic silicon wafers..I'm pretty pleased with the attempts tbh.
@OpenAI just solved an Erdős problem.
Paul Erdős was a prolific mathematician who famously offered cash prizes for problems he couldn’t solve. There are ~1,200 of them. The hard ones, like this have sat open for decades.
This is a big deal.
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.
This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.