Dev open-sources Usul, a C++ NN library that lets you express models as plain math
Exact-Contact-3837 · reddit · 2026-11-10
A Reddit developer released Usul, a C++ neural network library built around letting data scientists define models directly as math, without heavy frameworks like TensorFlow. It ships with built-in tensor support, auto-grad, back-propagation, and decent speed. The author stresses it isn't vibe-coded and politely bans vibecoded PRs.
The backstory: he taught himself numerical/analytical descent and backprop by hand on paper, solved XOR at scalar level, then learned matrix calculus and hand-wrote the XOR problem in matrix syntax with xtensor, manually backpropagating errors. He also frames the project as a detox from LLM dependency — instant LLM answers had started to erode his ability to absorb information. Repo and contribution ideas are on GitHub.
More from Research
- Stanford AI100 releases third report: 'AI Goes Mainstream' on the scaling era — StanfordHAI · 2026-11-11
- 26,040 agent runs and 100B tokens: FrontierHarness v1.1 ranks harnesses across 5 models — stuffyokodraws · 2026-11-11
- Perceptron open-sources EgoHome, a sub-second egocentric household-video benchmark — lukas_m_ziegler · 2026-11-11
- Epoch AI chart: AI cost falling ~13x a year, faster than any tech in history — sebkrier · 2026-11-11
- AI Trials Observatory launches: 93 RCTs on how AI affects mental health and beliefs — DG_Rand · 2026-11-11
- Raschka: open LLMs have abandoned full attention — hybrids and linear models now split 50/50 — rasbt · 2026-11-11