Recently, our team was deep in the weeds trying to improve a timeseries database’s performance.

When we built Rill, we made a bet that metrics – concepts like revenue, MAU, ROAS — are the core primitive for semantic layers. This blog post is a deep technical dive into how Rill's Metrics SQL works, why we built it the way we did, and where we're taking it next.

Rethinking AI power: why the public must govern the systems shaping our lives.

A Tesla owner crunched the numbers on how much he saved on fuel costs after 52,000 miles of driving a Tesla Model Y.

External GPUs now turbocharge Mac Minis for AI workloads

The French Revolution Digital Archive, a partnership between Stanford University and the Bibliothèque nationale de France, was announced this week with some 14,000 high-resolution images.

macOS Internals Deep Dive Building a Development Kernel Collection Configuring our Mac to boot the Development Kernel Collection Putting our machine to work! When did Apple grace us with this feature? Undoing our work for OS updates Closing Thoughts

In case you’re unaware, I’m not a developer. I’m actually an autistic catgirl annoyed by suboptimal use of computing power, and fixing that happens to involve programming. Crucially, it also includes discussing foundational technology with people behind the scenes, and apparently that makes me more aware of social aspects of this sphere. So, I have opinions about criticism of crates.io for supply-chain attacks. After a dozen similar articles, I have some select words to voice about why it’s off the mark.

The memory wall -- the widening gap between processor throughput and memory bandwidth -- has become the defining hardware constraint of the artificial intelligence era, now compounded by a structural NAND flash supply crisis driven by AI demand. We propose a post-transistor, pre-quantum memory architecture built on single-layer fluorographane (CF), in which the bistable covalent orientation of each fluorine atom relative to the sp3-hybridized carbon scaffold constitutes an intrinsic, radiation-hard binary degree of freedom. The C-F inversion barrier of ~4.6 eV (B3LYP-D3BJ/def2-TZVP, this work; verified transition state with one imaginary frequency; confirmed at 4.8 eV by DLPNO-CCSD(T)/def2-TZVP; rigorous lower bound from the fluorophenalane molecular model) yields a thermal bit-flip rate of ~10^{-65} s^{-1} and a quantum tunneling rate of ~10^{-76} s^{-1} at 300 K, simultaneously eliminating both spontaneous bit-loss mechanisms. The barrier lies below the C-F bond dissociation energy (5.6 eV) at both levels of theory, so the covalent bond remains intact throughout the inversion. A single 1 cm^2 sheet encodes 447 TB of non-volatile information at zero retention energy. Volumetric nanotape architectures extend this to 0.4-9 ZB/cm^3. We present a tiered read-write architecture progressing from scanning-probe validation (Tier 1, achievable with existing instrumentation) through near-field mid-infrared arrays (Tier 2) to a dual-face parallel configuration governed by a central controller, with a projected aggregate throughput of 25 PB/s at full Tier 2 array scale. A scanning-probe prototype already constitutes a functional non-volatile memory device with areal density exceeding all existing technologies by more than five orders of magnitude.