Tag
Top Kaggle scores on the ARC-AGI-3 benchmark jumped from 7% to 56% in just 30 days, with small local models in a harness beating average humans on a benchmark designed to demonstrate human cognitive superiority.
TraceML is a tool for trajectory-level analysis of ML research agents, comparing Codex agent runs with Kaggle Grandmaster notebooks and finding that human experts explore broader strategy sets while agents collapse into narrow loops. Distilling human research skills into a ~1,000-token planning prompt lifted Codex scores on five of seven competitions; the work will be presented at NeurIPS 2026.
Game Arena is an open platform for evaluating large language models through competitive games like Chess, Poker, and Werewolf, enabling dynamic assessment of strategic planning and robustness.
This paper proposes CLIC, a visual analytics system that compares coding behaviors of large language models through token-frequency analysis, providing insights for LLM selection and prompt engineering.
Crucible is an experimental neurosymbolic stack integrating Mamba2 with Z3 verification, JEPA for concept prediction, and HDC memory, designed for free Kaggle T4 sessions.
Meta's autonomous AI research system AIRA₃ placed 8th out of approximately 4,000 teams to win gold in a NVIDIA Kaggle competition to fine-tune a 30B Nemotron model, outperforming human competitors with access to the same tools.
This article praises open-source contributors like Unsloth and SGLang, who through optimizing tools and technologies enable ordinary people to fine-tune large language models such as Qwen3.8-27B on consumer-grade GPUs, lowering the barrier to AI development.
NVIDIA shares five lessons from over 5,000 Kagglers who fine-tuned reasoning models using LoRA adapters and synthetic chain-of-thought data in the Nemotron Model Reasoning Challenge, focusing on verifiable data, token budget, and infrastructure.
Google and Kaggle's 5-Day AI Agents Intensive Vibe Coding course attracted over 353,000 participants, with thousands of capstone projects submitted; the course remains available as a self-paced guide.
A comprehensive guide to mastering Kaggle data science competitions, covering techniques in machine learning, GenAI, and LLMs, updated with new chapters on time series and generative AI.
A controversial Kaggle competition result where the winning submission is accused of being nonsensical AI slop, sparking debate about review standards.
The tweet shares a compressed video version of the essence of the Google and Kaggle joint AI Agent 5-day special course, removing fluff and repetitive content, retaining core dry goods.
FarmGPT AI is a multi-agent AI farming assistant built by the author for a Kaggle course, featuring crop disease detection, chatbot, farm planner, market intelligence, and a command center.
A guide on using Kaggle's free dual Tesla T4 GPUs (32GB VRAM) to run large LLMs with massive context windows, covering multi-GPU parallelism strategies in llama.cpp.
NVIDIA open-sourced a Kaggle plugin that enables coding agents to execute end-to-end Kaggle competition workflows via natural language, including context gathering, kernel reproduction, submissions, and dataset management.
A Kaggle Grandmaster is offering a free course covering the entire machine learning process, including EDA, decision trees, gradient boosting, feature engineering, time series, and competition strategies.
Google Gemma团队正在赞助Kaggle上的1天黑客松活动，提供奖金支持，鼓励社区使用Gemma 4构建轻量级工具或推动AI创新。
Unsloth enables free fine-tuning of a 31B parameter multimodal model on Kaggle using 4-bit quantization, requiring only 22-24GB VRAM for local runs.
Free 5-day AI Agents course on Kaggle using Gemini, covering topics from introduction to deployment.
LongDS is a benchmark for evaluating AI agents on long-horizon, multi-turn data analysis tasks derived from Kaggle notebooks; experiments show best models only achieve 48% accuracy with significant drop over long turns.