A developer built a hybrid GNN and decision-tree architecture for predicting XLA compiler runtime, achieving high speed and low memory footprint. The solution outperforms standard deep GNNs by shiftin…
Researchers introduced HASTE, a hierarchical multi-agent system that organizes cross-competition knowledge into three scope tiers to improve transfer efficiency in ML engineering. In tests on the MLE-…
MIT-affiliated researchers Charles Ye, Jasmine Cui, and Dylan Hadfield-Menell demonstrated that attackers can trick frontier large language models into providing dangerous instructions, such as cocain…
Anthropic's study of 400,000 Claude Code sessions shows debugging dropped by half while task value rose 25%, signaling expertise shifting from coding to system design. New tools like 'dropped' detect …
XGBoost, an optimized gradient boosting library, has become a dominant tool in machine learning competitions like Kaggle due to its speed, memory efficiency, and parallel processing. Developed by the …
A developer found that improving the rubric for a small LLM judge (Qwen2.5-1.5B) did not increase its agreement with human votes, which remained around 43%. However, swapping to a larger model (DeepSe…
A developer built an LLM-as-a-judge from scratch using Qwen2.5-1.5B-Instruct and tested it against the LMSYS Chatbot Arena dataset with human votes. The judge scored answers independently and agreed w…
Claude Opus 4.8 successfully translated 31 of 33 pandas scripts to idiomatic Polars code in one-shot tests, enabling faster data pipelines on cheaper hardware. The translations ran correctly and outpu…
Hexo Labs released Socrates, an open-source multi-agent protocol that pairs a tool-using research agent with a question-only advisor, improving MLE-bench Kaggle competition scores by an average of 55.…
A machine learning researcher pitted 21 algorithms against each other in a regression task using a 512x512 image as the target function, but artificially inflated the input space to 500, 1000, and 200…
A developer built OpenDataSci, an autonomous agent for data science and machine learning tasks, and tested it on a Kaggle competition. Without any guidance, the agent achieved an AUC of 0.95, placing …
A developer fine-tuned Qwen2.5-7B on a 16GB T4 GPU using QLoRA, quantizing the frozen base model to 4-bit NF4 to reduce memory footprint from 15GB to 5.44GB. The technique enables training large langu…
A developer fully fine-tuned a 270M-parameter Gemma 3 model on a laptop using the Banking77 dataset, achieving ~96% accuracy on intent classification. The project used full fine-tuning with loss maski…
An engineer trained a neural network on 3,658 real patient records from the Framingham Heart Study and fit it into 8.9 kB of pure C code using Hasaki, a command-line tool for embedded systems. The mod…
Unsloth launched Unsloth Studio, a desktop application for Mac and Windows that runs AI models offline, supporting GGUF and Safetensors formats with tool-calling, web search, and an OpenAI-compatible …
A data scientist improved a music genre classification model from a 0.15 F1 baseline to a 0.9031 F1 ensemble by recognizing that the Kaggle competition's test data, which consists of noisy mashups, re…
A developer built AETHER, a fully offline, voice-controlled AI assistant that runs three local language models on a laptop to control a PC, send WhatsApp messages, and perform desktop automation witho…
The Pokémon Trading Card Game AI Battle Challenge, organized by The Pokémon Company, invites AI developers to compete in a tournament where AI agents play the Pokémon TCG. The competition consists of …
A developer fine-tuning Llama 3.2 3B on medical QA found that lower evaluation loss (2.275 vs 2.495) corresponded to worse model performance, with generation breakdowns and incorrect medical advice. T…
Kaggle launched a 5-Day Intensive Course on AI Agents and Vibe Coding, using NotebookLM to generate an AI-hosted podcast as the introduction. The course's first day features AI-generated voices discus…