Little Bear Labs’ Post

Retrieval augmented generation, or RAG, has emerged as a popular technique to fine-tune LLM models and ground their responses in specific data. However, when it comes to RAG systems, it's important to pay attention to how you break up your data, also known as chunking. We discuss some strategies and what works best: https://lnkd.in/eVqSfD-s

Weekend discussion on Boosted Tree. There’s a new Kaggle competition on Tabular Data to see if GBDT is still the king of the hill: https://lnkd.in/e-txaA2x There are pervasive unchecked issues with respect to ‘benign overfitting’ in Machine Learning in general where the performance of models are very good (the winner) with respect to testing data (based on random training-testing split) but these models are very sensitive to small perturbation in the inputs; thus, very fragile in production: https://lnkd.in/gYQ8_VNy In PiML, we put robustness test to avoid this issue: https://lnkd.in/e8Sp4jek Models should be selected not only on their performance but also their robustness. PiML GitHub: https://lnkd.in/dwn2bFng

Thanks for another insightful article, Agus!

https://buff.ly/3KNLs3i The Question of the Week from our free blog looks at Model or Algorithm? This is part 2 of a series. Algorithm rules are best followed explicitly, exactly, and consistently.  Model (which applies the algorithm) requires judgment and adaptation to the data at hand.

Learn more. Question your own expert process. Get with the times.

https://buff.ly/3VvmVoD The Question of the Week from our free blog looks at Model or Algorithm? This is part 2 of a series. Algorithm rules are best followed explicitly, exactly, and consistently. Model (which applies the algorithm) requires judgment and adaptation to the data at hand.

https://buff.ly/3RydvHW The Question of the Week from our free blog looks at Model or Algorithm? This is part 2 of a series. Algorithm rules are best followed explicitly, exactly, and consistently. Model (which applies the algorithm) requires judgment and adaptation to the data at hand.

https://buff.ly/3KOQoF7 The Question of the Week from our free blog looks at Model or Algorithm? This is part 2 of a series. Algorithm rules are best followed explicitly, exactly, and consistently. Model (which applies the algorithm) requires judgment and adaptation to the data at hand.

When working with unstructured data, there is no substitution for vector embeddings. Learn to use vectors in ML to locate and fix data errors, discover samples not present in your training data, and find hallucinations. 🔍 👉 https://bit.ly/46jIFIj #MachineLearning #VectorDatabase

https://buff.ly/4enmkOB The Question of the Week from our free blog looks at Model or Algorithm? This is part 2 of a series. Algorithm rules are best followed explicitly, exactly, and consistently. Model (which applies the algorithm) requires judgment and adaptation to the data at hand.