Robert Pillar-Smith

Looking for purpose, tracking the drift and making cool things

[Robert Pillar-Smith @rrps700]

New preprint: Exiting the Thompson Regression via the FM Operator

Passive Spectral Verification as a hardware trust primitive that uses an FM operator + a simple LC bandpass filter (capacitor, inductor, resistor) to verify silicon configuration with ZERO software in the verification path.

Exits the Thompson regression without software

https://t.co/iN0AYnpDh8

#HardwareSecurity #TrustingTrust #FPGA #CyberSecurity #HardwareTrust #Oscillators

[Robert Pillar-Smith @rrps700]

@lauriewired Preemptive meme is elite, but now I'm counting three separate objects in frame: you, Keith, and the C++ documentary (which I see has already started). Is this peak linguistic ambiguity or next-level self-objectification? Either way, 10/10 composition

[Robert Pillar-Smith @rrps700]

@lauriewired Probably that's like 8pm for me

[Robert Pillar-Smith @rrps700]

@lauriewired Does this mean we're getting a ram video soon 🧐

[Robert Pillar-Smith @rrps700]

Tailslayer clicked something. Using latency fingerprints to recover physical structure the OS is hiding. Genuinely clever inversion.
The GPU point is the key one. Tiles work because the geometry is exposed. That constraint isn't physical, it's architectural convention.
Curious what programming models look like when address space geometry matches compute geometry natively.

[Robert Pillar-Smith @rrps700]

@davepl1968 The spending time in the garden is only after you maintained the garden well enough that it looks good and you can have a sit down

[Robert Pillar-Smith @rrps700]

Hard problems aren't "hard"... they're hard *for a Turing machine on a Von Neumann substrate*.

New paper argues P vs NP is a philosophical category error: complexity is relational, not intrinsic. Substrate matters. Oscillators, memcomputing & phase-domain architectures change the cost landscape.

"On the Misclassification of Hard Problems"

https://t.co/u11YasL6K9

#PvsNP #UnconventionalComputing #Complexity

[Robert Pillar-Smith @rrps700]

@lauriewired Nice one you make good content that asks interesting questions.

[Robert Pillar-Smith @rrps700]

Can you ever really be too old to stare at an ant colony

[Robert Pillar-Smith @rrps700]

@lauriewired I wish I could afford to go. Hope you have a nice time in the UK I recommend tea and lots of it

[Robert Pillar-Smith @rrps700]

@lauriewired It's not worse-is-better. It's wrong in the right direction. The dirt path isn't worse than the switchback — it's wrong about being a path and right about where people want to go. C isn't worse than Lisp. It's wrong in a direction that compounds.

[Robert Pillar-Smith @rrps700]

@lauriewired Gabriel is pretty relatable I'm also a third rate computer scientist

[Dave W Plummer @davepl1968]

US Data Centers use about 150B gallons of water.
U.S. golf courses use about 500B gallons of water.
US Almond Farms use about 1500B gallons of water.

Now you know the scale of the problem. We cannot go on like this. We MUST control our almond consumption.

[Robert Pillar-Smith @rrps700]

@paleochristcon Most of the ones they tested in the factory were quoted as saying "please don't eat me"

[Robert Pillar-Smith @rrps700]

@elonmusk But I'm a jelly failure and want the money and free plasticine for some reason...

[Robert Pillar-Smith @rrps700]

Yes, fair point. With what is public so far I see exactly why that objection arises. I already have a clean substrate-level answer, and a rather satisfying one, but the full details stay under wraps for now. The direction is inferable from the latest paper even if I stopped short of claiming it outright on Zenodo.

[Robert Pillar-Smith @rrps700]

cool outputs from a project I'm working on
Navier-Stokes CFD running on the Orb42 v3 emulator (different grid resolutions). Ladder filter active keeping the vortex nice and coherent while the smoke blob evolves.

32x32 • 64×64 • 128×128 grids

#CFD #FluidSimulation #Emulator #HardwareProject #fpga

[Robert Pillar-Smith @rrps700]

Spoken claim in the clip: 12,000 arrests for online communications offences. Times FOI, cited in Hansard (Lords debate, 17 July 2025): 12,183 arrests in 2023 under s.127 Communications Act 2003 and s.1 Malicious Communications Act 1988. Roughly 33 a day. Of those, 1,119 sentenced, only a fraction custodial.

The on-screen "prison" was sloppy, granted. But "neither is it true" dodges the substance. People are going to jail for posts (threats, incitement, "grossly offensive" messages under s.127), and the arrest machine catches an order of magnitude more.

Arresting 33 people a day for messages, many of them non-violent, is the core issue worth debating. Not the chyron wording.

[Robert Pillar-Smith @rrps700]

Guri's work is great but it's a different problem... RAMBO exploits EM emissions from a system already assumed compromised. The Thompson regression is upstream of that: how do you trust the toolchain before the system runs at all? The LC filter exits the compiled software domain entirely rather than observing it from within

[Robert Pillar-Smith @rrps700]

@lauriewired Ooh I like the sound of that. I've recently been getting way too into FPGAs