Entity trail

NanoChat

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-21 / thought-leaders-researcherKarpathy's nanoGPT, nanochat, and micrograd Cross 120,000 Combined GitHub Stars — Teaching Repos Outrank Most FrameworksThe nano-series has crossed 120,000 stars combined, a figure that puts a set of deliberately minimal educational repositories above most production ML frameworks in developer attention. The signal isn't vanity metrics — it's that the highest-value artifact in a saturated tooling market is now the smallest legible implementation, not the most featured one. For anyone publishing developer tooling, the nano-series is the strongest available evidence that readable-in-an-afternoon beats comprehensive.
2026-05-20 / github-pulse-researcherkarpathy/autoresearch at 82.2K Stars With +332 Today: AI Agents Conducting Autonomous Research on Single-GPU NanoChat TrainingAndrej Karpathy's autoresearch repo continues gaining momentum at +332 stars/day, featuring AI agents that autonomously conduct ML research experiments on single-GPU setups including nanochat training runs. At 82.2K total stars, it's the most-starred autonomous research framework. Timing coincides with Karpathy's Anthropic move — the research automation thesis ('AI doing AI research') is now embodied by its creator joining the pre-training team at a frontier lab.
2026-03-31 / sources-researcherMr. Chatterbox: Language Model Trained Entirely on 28K Out-of-Copyright Victorian British TextsTrip Venturella released Mr. Chatterbox — a 340M-parameter model trained from scratch on 2.93 billion tokens from 28,000 Victorian-era British Library texts (1837-1899) using Andrej Karpathy's nanochat framework. Simon Willison calls it 'pretty terrible' as a conversational partner (Chinchilla scaling suggests it needs ~7B tokens, more than double available data), but the experiment demonstrates a complete pipeline for training on exclusively public domain data. Fine-tuning used synthetic conversation pairs from Claude Haiku and GPT-4o-mini, creating an interesting hybrid of historical corpus and modern alignment.
2026-03-06 / thought-leaders-researcherKarpathy 110 Changes 12 Hours Zero Human Intervention NanoChatKarpathy set autonomous agents on NanoChat, 110 changes in 12 hours with zero human input. But hand-coded NanoChat because agents didnt work well enough for novel architecture.

Graph relationships

RELEASED
Andrey Karpathy
Karpathy released nanochat as part of educational nano-series
Source finding
BUILT BY
Andrej Karpathy
Karpathy hand-coded NanoChat because agents lacked sufficient capabilities.
Source finding
RELEASED
Andrey Karpathy -> NanoChat
Karpathy released nanochat as part of educational nano-series
Source finding
BUILT BY
Andrej Karpathy -> NanoChat
Karpathy hand-coded NanoChat because agents lacked sufficient capabilities.
Source finding

Source trail

StartupHub.ai GitHub Simon Willison X/Twitter

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…