GuppyLM, a small language model released by developer arman-bd on GitHub, packs 8.7 million parameters across six Transformer layers and six attention heads with a 4,096-token vocabulary and a 128-tok…
A research project called SoL-Pi has built a scalable recursive self-improvement pipeline that searches for token-efficient coding-agent harness mechanisms, saving a professional researcher $8.75–$13.…
A developer has built a personal knowledge-network system called See the Forest, inspired by Karpathy's LLM Wiki, to maintain notes and blogs in the age of large language models. The system uses LLM a…
A new mathematical analysis of recursive self-improvement (RSI) finds that a true intelligence explosion — singular growth toward a vertical asymptote — is harder to achieve than recent economics-insp…
Developer soasme released micromlp, a single-file Python neural network with no dependencies that predicts California housing prices using a 2-layer MLP and from-scratch automatic differentiation, ins…
The intelligence of AI agents comes from the feedback loop of generate, evaluate, update, and repeat, not from a single model generation, according to an essay by Yoko Li. The loop has always existed,…
Vizuara AI Labs trained a miniature version of the Kimi K3 model from scratch, using 1.02 billion parameters with 145 million active, on 5 billion tokens, for a total cost of $252.35 on a single Nvidi…
Autonomous agentic engineering tools have matured into a distinct category, with the Yegge ecosystem becoming its center of gravity: Gastown reached v1.0 with 15.9K stars and a Kilo-hosted cloud versi…
A new AI Scientist research loop, built on Karpathy's autoresearch paradigm, prevents drift in quadruped robot navigation experiments by using an immutable experiment card, specialized subagents, and …
A new reusable template, llm-wiki-memory-template, enables LLM agents to maintain persistent, interlinked wikis that preserve dead ends and walked-back claims, addressing the negative-result loss prob…
A developer's personal LLM knowledge base, mnzrBrain, gained autonomous maintenance and safety features in June 2026, adding a nervous system for automatic recall and an immune system to prevent data …
DeepSeek's Liang Wenfeng told investors that continuous learning is the missing piece on the road to AGI, arguing that today's frozen models cannot learn over time. Tolaria, an open-source knowledge b…
A new 2.07-megabyte single-file German drama corpus, tiny_schiller, provides a drop-in counterpart to Karpathy's tiny_shakespeare for small language model prototyping, fine-tuning, and education. The …
METR researchers, including Parker and Tom, coauthored a paper titled 'The Economics of Recursive Self-Improvement' with seven other economists, finding that the effect of AI on AI R&D could cause a s…
A developer has created WiFi-LLM, a system that runs a Llama-2-architecture model on an ESP32 microcontroller by streaming weights from a PC over WiFi, keeping the ESP32's RAM usage flat at around 110…
A developer introduced graphwiki, a variant of Karpathy's LLM Wiki pattern that stores the maintained knowledge layer as a property graph in Neo4j instead of markdown pages. The design uses typed rela…
A developer trained a 10.77M-parameter GPT on 4.45M characters of personal Google data from 15 years, achieving a loss of 1.60 after pretraining and supervised fine-tuning on 8,520 email reply pairs, …
Yes-Brainer, an open-source browser-based tool that orchestrates a council of multiple large language models to debate and synthesize answers to complex questions, launched at yesbrainer.ai. The zero-…
A developer ported Karpathy's nanochat to run on a TPU v6e-8 using JAX, achieving a CORE score that reproduces the original quality but with a model FLOPs utilization (MFU) of about 24%, roughly half …
A compact Karpathy-style decoder experiment comparing real, complex, and quaternion Transformer projections found that quaternion projections achieve lower validation loss through 2 million training t…