DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining
In this paper, we propose Domain Reweighting with Minimax Optimization (DoReMi), which first trains a small proxy model using group distributionally robust optimization (Group DRO) over domains to pro...
OpenAI Execs Warn of Risk of Extinction in New Open Letter
Executives from OpenAI, DeepMind, and other AI researchers caution that their lifes work could potentially extinguish all of humanity. The Center for AI Safety released a statement about the risks of ...
Building Better Pangenomes to Improve Genomic Equity: An International Collaboration Announces Completion of First Human Pangenome Reference
Google AI Blog shares how an international collaboration of 119 scientists across 60 institutions, known as the Human Pangenome Research Consortium, has completed the first human pangenome reference. ...
Nvidias CEO Talks About the Future of Work: A.I. Wont Steal Jobs, but Skilled Experts Will Benefit
Nvidia’s CEO, Jensen Huang, discussed the transformation of the workplace due to Artificial Intelligence (A.I.) during his graduation speech at National Taiwan University. While A.I. won’t steal jobs,...
OpenAIs Short-Term Plans According to Sam Altman
Last week, Sam Altman sat down with 20 developers to discuss OpenAIs APIs and product plans. One major theme that emerged is that OpenAI is heavily GPU limited at present, slowing down short-term plan...
They Plugged GPT-4 Into Minecraft—and Unearthed New Potential for AI
An AI researcher at Nvidia led a team to set the GPT-4 model loose inside Minecraft, resulting in the creation of a Minecraft bot called Voyager that suggests goals, generates code, and refines its ac...
Voyager: An Open-Ended Embodied Agent with Large Language Models
Voyager is an embodied lifelong learning agent in Minecraft that utilizes large language models for skill acquisition and exploration. It consists of three key components: an automatic curriculum that...
Tree of Thoughts - An All-New Powerful and Flexible Algorithm to Advance Model Reasoning by 70%
# Tree of Thoughts 🌳🌲🌴🌿🍃 Tree of Thoughts (ToT) is an all-new powerful and flexible algorithm that advances model reasoning by a whopping 70%. This is a plug-in-and-play version, connect your ow...
State of GPT | BRK216HFS
Andre Carpathy, an AI researcher and founding member of OpenAI, delivered a talk on the state of GBT and the evolving ecosystem of large language models. The talk consisted of two parts, covering GBT ...
RWKV: Reinventing RNNs for the Transformer Era
Transformers have revolutionized almost all natural language processing (NLP) tasks but suffer from memory and computational complexity. In contrast, recurrent neural networks (RNNs) exhibit linear sc...
Governance of Superintelligence: Managing Risks to Ensure a Prosperous Future
As AI systems become more capable, they will soon exceed expert skill levels in most domains and carry out as much productive activity as one of todays largest corporations. This calls for special tre...
Tree of Thoughts: Deliberate Problem Solving with Large Language Models
Language models are being deployed for general problem solving, but can fall short in tasks requiring exploration or strategic lookahead. To address this, a new framework called Tree of Thoughts (ToT)...
Alan Turing Institutes Failure in Developing Modern AI in the UK
The Alan Turing Institute and the AI council have failed to acknowledge the importance of large language models (LLMs) in the development of modern AI in the UK. Despite its flagship status, the insti...
Improve Language Model Performance with Guidance Programs
Guidance programs improve the control of language models for better performance. They allow for interleaving generation, prompting, and logical control for a continuous flow. Using powerful language m...
LIVE: OpenAI CEO Sam Altman testifies during Senate hearing on AI oversight — 05/16/23
The Senate Judiciary Committees Subcommittee on Privacy, Technology and the Law held a hearing on the oversight of artificial intelligence (AI). While AI advancements can bring positive changes such a...
AI Test Kitchen - A place to experience and give feedback on Googles latest AI technologies
AI Test Kitchen is a platform that allows people to experience and provide feedback on Googles latest AI technologies. The goal of this initiative is to learn, improve, and innovate responsibly on AI ...
OpenAI Peeks into the “Black Box” of Neural Networks with New Research
In a new research paper, OpenAI details a technique that uses its GPT-4 language model to write explanations for the behavior of neurons in its older GPT-2 model. This step forward for interpretabilit...
Case Study of Future Me Campaign: Using Artificial Intelligence to Visualize Retirement
In this case study, Michał Bucholc explores how ING Bank Poland used the power of artificial intelligence to encourage clients to think about their financial future and visualize retirement. The Futur...
Google Keynote (Google I/O ‘23)
Google CEO Sundar Pichai opened the companys annual I/O developer conference by outlining how AI is helping to drive the tech giants future. Pichai said that generative AI is redefining Googles core p...
Google Launches AI-Centric Coding Tools, Including Competitor to GitHub’s Copilot
At its annual I/O developer conference, Google launched a variety of AI-centric coding tools, including its own version of GitHub’s Copilot. Google introduced a chat tool to ask questions about coding...