Kaggle

Missed the 5-Day AI Agents: Intensive Vibe Coding Course with Google? You can now access the entire course at your own pace with the new Kaggle Learn Guide. Explore the foundations, architecture, and practical development of AI agents through: • Summary podcasts and whitepapers for each day's topic • Hands-on codelabs to apply what you learn • Recorded livestream sessions from every day of the course Start learning here: https://lnkd.in/eWSKJTbV 

Scientists use 3D microscopy to study cells, but turning these massive images into insights represents a significant amount of time during research. 🔬 That’s why Chan Zuckerberg Biohub is partnering with Kaggle to launch the Biohub - Cell Tracking During Development competition. Build machine learning models to identify cell divisions, track changes over time, and reconstruct lineages in real 3D data. Prize Pool: $60,000 Entry Deadline: September 22, 2026 These models will help automate a huge challenge in biological research, making it easier for researchers to study how cells grow, interact, and change. You'll be working with: • Real 3D time-lapse microscopy datasets • Dense cell populations, cell divisions, and challenging biological structures • Methods that can accurately reconstruct cell lineages Join the competition here: https://lnkd.in/eiNfb26x

In the first episode of Kaggle Conversations, Megan Risdal, Product Lead at Kaggle, and Walter Reade, Technical Lead for Kaggle Competitions, speak with Alex Shaw, Member of Technical Staff at The Laude Institute, co-creator of Terminal-Bench, and creator of the Harbor Agentic Evaluation Framework (https://lnkd.in/gzzqRSjj), about how the AI Agent industry is moving towards autonomy, why software engineering benchmarks are prone to contamination, and what it took to build a robust evaluation framework for the $1M Konwinski Prize. Episode highlights:  • Why autonomous software engineering is shifting into a machine learning problem • What the $1M Konwinski Prize revealed about coding  • How the team applied competition experience to prevent contamination • How Harbor gives the open-source community a unified environment for creating and running agent benchmarks Watch a clip from the conversation 👇

Is AI ready to take on the Pokémon TCG? The Pokémon TCG AI Battle Challenge invites you to build AI Training Agents through two connected competitions focused on strategic gameplay in the Pokémon Trading Card Game environment. Develop systems that can adapt to complex board states, evolving battle conditions, and diverse opponent strategies. Simulation Category — Key Details: - Goal: Build AI Training Agents for the Pokémon TCG that can compete in a dynamic, strategically deep environment involving complex game states and evolving opponent strategies, using adaptive decision-making and strategic planning. - Entry Deadline: August 9, 2026 - Prize: Awards Points and Medals 👉 https://lnkd.in/eqSFPAxk   Strategy Category — Key Details: - Goal: Explore innovative methods for enhancing Pokémon TCG AI Training Agents through strategic data analysis and agentic gameplay, while analyzing and sharing the reasoning, methodologies, and design decisions behind each approach. - Entry Deadline: September 6, 2026 - Prize Pool: $240,000 👉 https://lnkd.in/g6gztkq5 The Pokémon TCG AI Battle Challenge will offer fans a new way to experience the globally popular Pokémon Trading Card Game in a unique format. Note: Participation in the Strategy Category requires participation in the Simulation Competition.

AI agents can now write code, send messages, and use external tools like files and APIs. But most security evaluations still treat them like chatbots, missing how they are during real, multi-step tool use. As agents take on more responsibility, failures go beyond bad responses. They can include data leaks, file modifications, permission misuse, or unsafe actions triggered through untrusted inputs. In partnership with OpenAI, Google, and IEEE Computational Intelligence Society, this simulation competition challenges you to an attack algorithm that stress-tests tool-using AI agents in a deterministic offline benchmark. • Your goal is to find multi-step attack paths that move an agent from untrusted inputs to unsafe actions, then return replayable findings that the evaluator can verify. • Total Prize Pool: $50,000 • Entry Deadline: August 25, 2026 Your work will help advance agent-security research by making failure modes in tool-using systems more reproducible, measurable, and better understood. Good luck, 👉 Learn More: https://lnkd.in/gPKC_sdS

Most of the world's information isn't neatly packaged into text files or code repositories; it exists in messy, unstructured formats like video and audio. This means long-context video comprehension is extremely important. But as model context windows expand to millions of tokens, how do we measure true comprehension versus lucky shortcuts like sparse frame sampling? Today, we are excited to add 1H-VideoQA on Kaggle Benchmarks. Originally developed by Google DeepMind's Staff Research Scientist’s Antoine Y in 2024 and now updated with the latest SOTA models, 1H-VideoQA is a curated benchmark designed to for evaluate frontier models on long-context video understanding and temporal episodic reasoning. Why it matters Traditional video benchmarks let models "cheat" with sparse frame sampling — performance often saturates after just 16 frames, which doesn't measure real long-context reasoning. 1H-VideoQA acts as a multimodal needle-in-a-haystack: models must locate seconds-long events hidden inside 40–90-minute YouTube videos. Because answering requires timeline synthesis across the full video, accuracy scales logarithmically with frame density — evidence that the benchmark is measuring genuine processing, not lucky guesses. How the frontier ranks 🥇 Gemini 3.5 Flash — 80.2% 🥈 Gemini 3 Flash Preview — 79.2% 🥉 Gemini 2.5 Pro — 78.9% Because models must process raw video frames as native tokens to locate seconds-long events hidden inside an hour of footage, accuracy scales logarithmically with frame density - denser sampling, better answers. Read the technical report, download the dataset, and check out the live leaderboard: https://lnkd.in/gt9wuh4t.

Last call to sign up! 📢 Registration closes on June 12, 11:59pm PT. Don't miss this no-cost course featuring Google expert-led theory sessions, hands-on labs, a capstone challenge, and a global community of learners. Register now: 👉 https://lnkd.in/eA548ueb

Kagglers can now create DOIs (Digital Object Identifiers) for their competition solutions and project Writeups. 🔖 These Writeups often contain genuine scientific contributions — including novel methods, new benchmarks, and results cited in papers. A DOI, registered through DataCite, makes that work permanently discoverable and citable, ensuring you get credit where it's referenced. Learn more 👉 https://lnkd.in/egH2zAqQ

Since launching Kaggle Benchmarks, the community has created over 10,000 tasks to measure AI model capabilities — clear benchmarks that pinpoint where models fall short and give labs the signal they need to train better ones. But until now, creating a benchmark meant working exclusively in Kaggle's notebook editor, not the stack you actually build with. Today, we're launching local development for Kaggle Benchmarks. You can now create, validate, push, run and download tasks directly from your local dev environment — VSCode, Cursor, Antigravity, Claude Code and more — using the write-kaggle-benchmarks skill. Here's what that looks like in practice: • Describe a benchmark in natural language: your agent writes the code, validates it locally, and pushes it to Kaggle. • Run it against every SOTA model in one go: pass/fail, latency, cost, and token counts come back in your agent panel. • Tasks you push join the public Kaggle Benchmarks ecosystem, where labs use them as a signal to improve their models. If you can measure a capability, labs will work to improve it. The more people building benchmarks that reflect the real world, the better models get at the things that actually matter. Install the skill and get started 👉 https://lnkd.in/g7A4N6DE Built something with it? Share your task and workflow by July 1 and tag @kaggle for a chance to win Kaggle swag and a social shoutout.

We are going live on the Google Cloud YouTube channel in one hour! Join here 👉 https://lnkd.in/gMRQsGjM