Arun Jagota·12h agoGenerative World ModelsA generative world model gives one a simulated world to act in. It can imagine what happens next given an action in this world. For…
Arun Jagota·3d agoMulti-Modal Large Language ModelsConsider the image below and the prompt: find me a path from the maze’s entrance to its exit.
Arun Jagota·5d agoReinforcement Learning With Automatically Verifiable RewardsTraditional Reinforcement Learning with Human Feedback requires a human in the loop. While this is useful for alignment and safety, this…
Arun Jagota·Nov 5Reasoning In Large Language ModelsLarge language models (LLMs) predict the next word in a response. Reasoning systems, built on top of LLMs, predict the next logical step in…A response icon1A response icon1
Arun Jagota·Nov 5Automated Feature EngineeringIn real-world regression or classification problems, the input may need to be suitably transformed to make the target easier to predict…
Arun Jagota·Oct 29Transformer for ImagesThe transformer is hugely successful in modeling text. Via self-attention, it is able to discern relationships between words far apart…
Arun Jagota·Oct 28Learnable Self-Attention With Examples and AnalogiesOn the task of predicting the next word
Arun Jagota·Oct 18Basic Mechanisms in LLMs for Code Generation IllustratedConsider this partial Python code.