rekonquest.bsky.social

@rekonquest.bsky.social

9 Followers 9 Following

Just a guy with geometric intuition and too many ideas. Ideas form faster than I can build them.

Infographic titled "I built an AI that is not allowed to guess." A light cone diagram labeled "the causal structure of spacetime, applied to AI" shows a purple future cone (what it can affect), a blue past cone (what could have caused it), and a white center dot marked "event". Outside the cone, in the region labeled "spacelike: unreachable", sit three red X marks: hallucinations, prompt injection, and hacking. Text blocks explain: anything from the outside is treated like a future event that does not exist in its world and is refused; the system uses cognitive primitives with tuned chaos and shows emergent behavior with zero ML; a 4B model only translates between the user and the core system; the goal is no weights and no training data, storing information the way the universe stores it. Five stat boxes read: 4/4 laws learned by the adaptive governor, 0/4 for the fixed schedule on the same stream, 40/40 incoherent noise rejected, 1.25 exact conserved chaos budget, and 272 tests passing with zero errors. The footer reads: "The cone decides. Not the vibes." with the link github.com/Lucerna-Labs/atom-harness.
Promotional card for "Atom Telemetry Blocker", an open-source Rust privacy tool for Windows. Top labels: "Open Source" (purple), "Rust" (amber), "Windows" (green), "Privacy Tool" (gray). Large title: "ATOM TELEMETRY BLOCKER" with "BLOCKER" highlighted in purple, followed by a horizontal rule. Tagline: "Garble the data, keep the encryption, break the tracking." Body text explains that Microsoft's telemetry services read registry values to build a permanent device fingerprint (the GDID), and that the tool's chaos-high mode randomizes those values on a 500ms cycle, so the encrypted HTTPS still flows but decrypts to random noise. Footer: GitHub URL "github.com/Lucerna-Labs/atom-telemetry-blocker" with the attribution "by Lucerna Labs · Rust + Windows + chaos." A small 6x6 grid in the upper-right corner visualizes the contrast: the left half is uniform gray ("telemetry"), the right half is randomized purple, red, and amber ("chaos").
Simple cream-colored portrait graphic with a narrow teal stripe on the left. Header: Cognitive Primitives in a Functional Database. Large title: AtomDB. Text reads: This is AtomDB. A fully functional hybrid database built in Rust from mathematical, retrieval, graph, optimisation, and cognitive primitives, including deterministic hashing, n-gram indexing, BM25, PPMI co-occurrence, fuzzy vocabulary matching, graph walks, outcome learning, counterfactual imagination, and synthetic control. It handles storage, sparse retrieval, vector search, graph relationships, learning, and planning inside one engine. Its retrieval system cuts LLM context and token usage by more than 95%, while reaching near-ML retrieval quality in the top 3 to 5 results and near-perfect retrieval on coding tasks. Most AI systems waste context and money by throwing large amounts of mostly irrelevant data at the model. AtomDB retrieves the small amount that actually matters and uses cognitive primitives to learn from outcomes, model alternatives, and improve future retrieval decisions.
an image that my 3D engine produced