Provide personalized recommendations without knowing your users. — Each recommendation system is different, and some of them are much easier to build than others. Think about Netflix. They know all about each of their movies, have rich personal user data, and an abundance of user-produced data: plays, ratings, watch time, and so on. In this data-rich environment, one…

Streamline your ML workflow management — Have you ever copy-pasted chunks of utility code between projects, resulting in multiple versions of the same code living in different repositories? Or, perhaps, you had to make pull requests to tens of projects after the name of the GCP bucket in which you store your data was updated? Situations…

Meta’s two new models learn more as humans do, and what do small language models actually learn? — In this edition: ImageBind from Meta, a model learning from six modalities; I-JEPA, the first step towards Yann LeCun’s dream of human-like AI; Will Microsoft’s Orca set the trend for small language models? AI Pulse is also available as a free newsletter on Substack.

Hint: forget about the p-values — Data drift is a concern to anyone with a machine learning model serving live predictions. The world changes, and as the consumers’ tastes or demographics shift, the model starts receiving feature values different from what it has seen in training, which may result in unexpected outputs. Detecting feature drift appears…

Are vision foundational models catching up with LLMs? — Self-supervised training methods continue to deliver breakthrough after breakthrough. Last week, Meta AI released the second version of their self-DIstillation with NO labels or DINO model. The model can supposedly be used as a backbone to solve virtually any computer vision task without fine-tuning! Have the foundational models in computer…

A new foundational model for computer vision, making sense of the spree of open-source LLMs, and should AI be open-source? — AI Pulse is also available at pulseofai.substack.com. In this edition: DINOv2, a universal computer vision backbone; A spree of open-source LLMs emerges following LlaMa’s leak; Should AI models be open-sourced?

Reduce your models' latency, storage, and inference costs with quantization and pruning — Over the last few years, machine learning models have seen two seemingly opposing trends. On the one hand, the models tend to get bigger and bigger, culminating in what’s all the rage these days: the large language models. Nvidia’s Megatron-Turing Natural Language Generation model has 530 billion parameters! …

Bard, Sparrow, and multimodal chatbots will render it obsolete soon, and here is why. — ChatGPT, the OpenAI chatbot released last autumn, has taken the Internet by storm. Arguably, no other machine learning model has ever made so many headlines outside of the AI community. It provides a near-human experience when talked to and can help many of us do our work faster: from SEO…

You don’t need a perfect oracle to get your probabilities right. — Machine learning models are often evaluated based on their performance, measured by how close some metric is to zero or one (depending on the metric) but this is not the only factor that determines their usefulness. In some cases, a model that is not very accurate overall can still be…