We are launching a new app called Sora. This is a combination of a new model called Sora 2, and a new product that makes it easy to create, share, and view videos. This feels to many of us like the “ChatGPT for creativity” moment, and it feels fun and new. There is something great about making it really easy and fast to go from idea to result, and the new social dynamics that emerge. Creativity could be […]
Science in the age of foundation models To transform scientific domains, foundation models will require physical-constraint satisfaction, uncertainty quantification, and specialized forecasting techniques that overcome data scarcity while maintaining scientific rigor. Machine learning Danielle Maddix Robinson September 26, 10:58 AM September 26, 10:58 AM Deep learnings transformative impact has become widely evident with the advent and rapid adoption of large language models (LLMs). LLMs and other so-called foundation models are potentially powerful tools for advancing a wide range […]
TL;DR Grounding augments the pre-trained knowledge of Large Language Models (LLMs) by providing relevant external knowledge along with the task prompt. Retrieval-augmented generation (RAG), which builds decades-long work in information retrieval, is the leading grounding strategy. The key challenges in LLM grounding revolve around data. It has to be relevant to the task, available in the right quantity, and prepared in the right format. When providing information to an LLM, less is more. Research shows that it is […]
Introduction In this article, we will explore the details of RESTful architecture, the merits of Fast API, and how to create a simple API with Fast API. But before that, we shall discuss some basics of APIs like their functionalities, uses, and architectures, and then we will move on to more exciting stuff related to REST and Fast API. What are APIs? Imagine you went to a restaurant with your partner for a dinner date, and you want to place […]
Growth in the use of AI services has been astonishing; we expect it to be even more astonishing going forward. As AI gets smarter, access to AI will be a fundamental driver of the economy, and maybe eventually something we consider a fundamental human right. Almost everyone will want more AI working on their behalf. To be able to deliver what the world needs—for inference compute to run these models, and for training compute to keep making them […]
TL;DR Instruction fine-tuning (IFT) refines pre-trained large language models (LLMs) to follow specific task instructions by training on prompt-response pairs. At the core of IFT is a dual-objective loss function that balances instruction-following with general language modeling capabilities. Each IFT training sample consists of a task, a context, and a target response. Datasets can be augmented through automated approaches to increase task diversity and difficulty. Modifications to an LLM’s input layer, attention mechanism, and output layer improve instruction-following […]
Scientific frontiers of agentic AI The language AI agents might speak, sharing context without compromising privacy, modeling agentic negotiations, and understanding users commonsense policies are some of the open scientific questions that researchers in agentic AI will need to grapple with. Conversational AI Michael Kearns September 11, 10:30 AM September 17, 10:19 AM It feels as though weve barely absorbed the rapid development and adoption of generative AI technologies such as large language models (LLMs) before the next […]
The webpage for my AI safety course is on https://boazbk.github.io/mltheoryseminar/ including homework zero, video of first lecture and slides. Future blog posts related to this course will be posted on lesswrong since many people interested in AI safety visit that. Video of first lecture Some snapshots from slides:
Introduction The area of machine learning (ML) is rapidly expanding and has applications across many different sectors. Keeping track of machine learning experiments using MLflow and managing the trials required to construct them gets harder as they get more complicated. This can result in many problems for data scientists, such as: Loss or duplication of experiments: Keeping track of all the many experiments conducted can be challenging, which increases the risk of experiment loss or duplication. Reproducibility of results: […]
AI has gotten remarkably better in recent years; ChatGPT can do amazing things that we take for granted. This is as it should be, and is the story of human progress. But behind the blinking circle, nicely abstracted away, is the greatest story of human ingenuity I have ever seen. A lot of people have worked unbelievably hard to discover how to build something that most experts thought was impossible on this timeframe, and to build a company […]