AI / Models / Agents

Soofi S β€” Sovereign German-English Foundation Model β€” Soofi Project’s pretraining report for a German–English foundation model.

Learning from experience instead of curated datasets Β· Oak Lab β€” a discussion of how agents can learn from experience rather than curated datasets.

Inkling: Our Open-Weights Model β€” an open-weights model released by Thinking Machines Lab.

Kimi K3: A New Frontier in Intelligence β€” an introduction and reading of Kimi K3.

A four-hour interview with Physical Intelligence’s Ke Liyiming β€” a conversation about PI’s open-model research, the evolution of robotics, and its major participants.

Have Chinese AI Models Caught Up to the US Frontier? β€” uses benchmarks and capability-gap estimates to ask whether Chinese models have caught the US frontier.

Agent Swarms and the New Economics of Models β€” Cursor’s discussion of agent swarms and their model economics.

From Speaking to Creating: Seed Audio 1.0 β€” ByteDance Seed Audio 1.0, an audio creation model release note.

[2607.10252] One Token Is Enough: Fingerprinting and Verifying Large Language Models from Single-Token Output Distributions β€” fingerprints and verifies LLMs from single-token output distributions, useful when the serving chain is opaque.

Introducing Laguna S 2.1 β€” Poolside’s Laguna S 2.1 release, emphasizing longer-horizon work and more effective use of reasoning.

Macaron-V1-Preview: 749B MoL Agent Model post-trained from GLM5.1 β€” a 749B MoL agent model preview post-trained from GLM5.1.

[2507.09369] A Taxonomy of Omnicidal Futures Involving Artificial Intelligence β€” a taxonomy and examples of AI-related futures in which nearly all humans are killed; framed as possibilities, not inevitabilities. image.png

Scaling Video Pretraining with Imagination Models β€” Induction Labs pretrained Photon-1 from scratch on 575 million frames of computer-use data, using autoregressive training in latent space.

[2607.20548] SOAP, Muon, and Beyond: Pushing LLM Pretraining Scales Β· NVIDIA-NeMo/Emerging-Optimizers β€” adapts and strengthens higher-order optimizers such as SOAP and Muon for large-scale pretraining, with related NVIDIA NeMo implementations.

How to train a frontier-level world model Β· next-state/open-dreamer β€” an open-source Dreamer implementation in JAX.

Causal Video Models Are Data-Efficient Robot Policy Learners β€” Rhoda AI’s DVA reformulates robot policies as video generation for better data efficiency and long-context behavior.

What is the purpose of interpretability? β€” asks what interpretability research is for and what problems it should serve.

Mathematics / Research / Learning

DSPIllustrations.com β€” interactive illustrations for understanding digital signal processing.

Wireless Pi β€” learning material on wireless communications and signal processing.

intro-to-dl β€” an introductory deep-learning course.

[1708.09093] Memories of a Theoretical Physicist β€” Joe Polchinski’s memoir-like notes as a theoretical physicist.

[2607.11664] Lecture Notes: Convex Optimization β€” lecture notes from a Graz University of Technology convex-optimization course, covering preliminaries, existence, and algorithmic basics.

Discussion of a three-dimensional map related to the Jacobian Conjecture β€” the original post gives a map from $\mathbb{C}^3$ to $\mathbb{C}^3$ and claims it is a counterexample; the conclusion should still be judged against a formal paper and peer review.

PyTorch internals β€” a guide to PyTorch’s internal mechanisms.

OpenResearch β€” an entry point for research resources and tools.

Byrne’s Euclid β€” an online edition of Oliver Byrne’s colorful rendering of Euclid’s Elements.

An introduction to the 2026 Fields Medals β€” a discussion of laureates Yu Deng and Hong Wang and their work.

Python Tutorials – Real Python β€” high-quality Python tutorials and practical articles.

[2607.20732] librla: Randomized Linear Algebra Library β€” a randomized linear-algebra library aimed at intermediate-sized matrices (up to roughly $10^4$ dimensions).

Programming / Systems / AI Tools

Synara β€” a development tool integrating Claude Code, Codex, Gemini, and Cursor.

ChatCut ChatGPT/Codex Plugin β€” a ChatGPT / Codex plugin for trimming and organizing conversation context.

xai-org/grok-build β€” an open-source coding-agent harness and full-screen TUI.

ZacharyZhang-NY/Kigi-CLI β€” an unofficial Rust community build of Kimi Code CLI based on Grok Build.

tldraw offline Β· nn-digits.tldraw example β€” an AI-operable desktop whiteboard.

petersolopov/yace β€” a browser code editor with plugin support in under 2 KB.

ayghri/i-have-adhd β€” an Agent Skill that constrains Claude Code’s output and brings the answer forward.

mesamirh/MovieBox-Tui β€” a MovieBox terminal interface written in Rust.

ChatGPT Table of Contents β€” a browser extension that generates a table of contents for ChatGPT conversations.

marcelroed/gigatoken β€” a language-model tokenizer designed for GB/s throughput.

Tangent β€” a browser centered on workflows and spatial organization.

Why I chose Tauri, and what it cost me β€” Tabularis on the experience and cost of choosing Tauri.

CaliCastle/skills β€” its Signal Geometry Skill turns one concept into a sparse abstract illustration or poster through precise geometry, low-contrast matte fields, and restrained spatial relationships. image.png

nexu-io/codex-slides β€” an open-source AI slide studio inside Codex with support for research, outlining, rendering, editing, presenting, and export.

Minibase β€” converts web pages to Markdown for Claude, ChatGPT, and Obsidian.

l0ng-ai/tty7 β€” a terminal workbench built in Rust with Zed’s GPUI, integrating shells, persistent sessions, SSH, and coding agents; the project reports roughly twice the throughput of Alacritty, Ghostty, or Kitty.

Design / Writing / Creative Work

NameThat β€” a visual dictionary of UI elements, useful for identifying unfamiliar patterns and describing design needs to AI more precisely.

Arjun Kocher β€” a portfolio website.

Making Software: How to make a font β€” a tutorial on making a font.

Vercel Design System Β· Dark β€” a complete set of guidelines for color tokens, typography, spacing, radii, components, motion, and writing. Its central principle is that color should communicate, not merely decorate.

Feldar Β· Marker Β· Ornata Β· Ishmael β€” apps focused on concentration, thinking, or long-form writing.

Attune β€” a personal computing / writing app oriented toward focus and rhythm.

pixgel β€” a liquid pixel-art editor with smooth SVG export.

blakeshao/a-psychos-gd-tool β€” a node-based graphic-design tool.

You’re relatively right! β€” a heatmap of writing-style similarity across LLMs, built from wording alone. QQ_1784653759506.png

Sticker Forge β€” an interactive sticker maker.

Canvas UI β€” a creative-canvas and WebGL component library.

Design inspiration and reference indexes:

News / Life / Other

θ΅›εšιΈ‘θ›‹ β€” digital products, deals, and offers.

Electroagenda β€” an electronic-music event calendar.

jobleap β€” an online job-application information tool.

Eli Bendersky’s website β€” writing on programming, computer science, and technology.

Scenic Relaxation Β· Jacob Adler Β· Relaxation Film β€” channels featuring landscapes and relaxation films.

There Is No Avenue of Communication Between You and God β€” an essay about the absence of a direct channel of communication between people and God.