iiinova — where your digital self keeps evolving

Weights & Biases

@weightsbiases.bsky.social

833 followers 228 following 13 posts

The AI developer platform.

Posts Replies Media

Weights & Biases @weightsbiases.bsky.social · 08/07/2026

Full write-up, all 677 features and how it cleared the gold line utm.io/urkYG

wandb.ai

Tabular ML didn't die. It was just waiting for agents

Learn how our tabular data agent, took #1 on the MLE-bench tabular slice with two golds, three above-median, four-for-four valid submissions. .

000

Weights & Biases @weightsbiases.bsky.social · 08/07/2026

None of this needed a smarter model. A 2018 XGBoost and a 2025 one score about the same. What never scaled was the engineer running 20 feature loops a week. The agent runs 60 overnight for pocket change, and trying things was most of the job anyway.

100

Weights & Biases @weightsbiases.bsky.social · 08/07/2026

it didn't get there by brute force. Nobody told it crystals are periodic. It worked out that raw atom-to-atom distances misread the chemistry across a cell boundary, then rebuilt them as metal-to-oxygen nearest neighbors. 677 features by the end. 😳

100

Weights & Biases @weightsbiases.bsky.social · 08/07/2026

The headline run was a materials-science task. Predict formation energy and bandgap of a transparent conductor from its crystal structure. It landed 1st of 685. OOF score 0.04853 against a 0.05589 gold line. The humans who won it originally spent weeks.

100

Weights & Biases @weightsbiases.bsky.social · 08/07/2026

Sixty cycles later, across four MLE-bench datasets at once, it had two golds and three of four finishes above the Kaggle median. No human in the loop the whole run. One NVIDIA L40 on @CoreWeave and a few dollars.

100

Weights & Biases @weightsbiases.bsky.social · 08/07/2026

Tabular ML didn't die. It was just waiting for agents. We rented one GPU for the price of a coffee and asked an agent to top a Kaggle leaderboard on its own. It came back with two gold medals. Here's the run 👇

100

Weights & Biases @weightsbiases.bsky.social · 18/12/2024

Ready to test these new features? Add your keys, start generating trials, and unlock the full potential of the playground.

080

Weights & Biases @weightsbiases.bsky.social · 18/12/2024

While trials work seamlessly with most models, some like o1 currently don’t support choices. We’re actively working to enable this in the future. Stay tuned!

170

Weights & Biases @weightsbiases.bsky.social · 18/12/2024

The magic doesn’t stop there. Once you pick the best output, you can continue your exploration as if that output is how the model would’ve answered. See how the conversation unfolds from your chosen result!

160

Weights & Biases @weightsbiases.bsky.social · 18/12/2024

Trials really shines when temperature settings are turned up. 🔥 Explore the model’s creativity by generating multiple outputs at once and comparing the diverse responses—all at a glance.

100

Weights & Biases @weightsbiases.bsky.social · 18/12/2024

Choosing the best model output often means iteration. Playground trials will save you time by letting you compare multiple results side-by-side before committing to one. 🕵

100

Weights & Biases @weightsbiases.bsky.social · 18/12/2024

We’ve added trials to the playground! Now, you can generate multiple outputs for the same input. Simply bump up the “Number of Trials” in the settings sidebar to see various options before continuing your work.

100

Weights & Biases @weightsbiases.bsky.social · 18/12/2024

🛠️ New tools in the W&B playground are here! Amazon Nova models & xAI's Grok beta LLM are now available. Add your API keys in team settings to start exploring these powerful models today. 🚀 Here’s what’s new: 👇

240