The Federal Trade Commission and Justice Department have reached a deal that would set the stage for antitrust probes into Microsoft, OpenAI and Nvidia, setting up unprecedented regulatory scrutiny of the companies' conduct in the AI race, according to a person familiar with the matter, who spoke on the condition of anonymity to discuss a probe whose details are not public.

Google first launched NotebookLM, its AI-powered research tool, in summer 2023 and now the company is expanding its capabilities and access. In a new blog post, Google announced that NotebookLM is officially running on Gemini 1.5 Pro, an upgrade from the original Gemini Pro model that brings a host of new features. The research and writing assistant has also expanded to over 200 countries and territories. NoteBookLM works differently than many other AI-powered tools in that it only takes from the resources you provide, rather than a plethora of general knowledge. The research assistant has always accepted files such as transcripts, research notes and corporate documents from Google Docs or in PDF format, but now you can also upload Google Slides and web URLs.

Microsoft is under investigation by the Federal Trade Commission over its deal with Inflection AI, according to The Wall Street Journal. Back in March, the company hired almost all of Inflection AI's employees, including founders Karén Simonyan and Mustafa Suleyman, who was also a DeepMind cofounder. In addition, Microsoft paid Inflection AI 650 million to license its artificial intelligence technology. Now, the FTC wants to know whether the companies deliberately structured the deal to avoid being the subject of regulatory antitrust review. As The Journal notes, companies are required to report any acquisition that's valued at 119 million or more to federal antitrust agencies.

This one, called AI Chat, comes courtesy of DuckDuckGo, the privacy-focused search engine that obviously doesn't want to feel left behind in the AI arms race. The company has been testing AI Chat over the last few months, but as of today, it's available to everyone. Unlike other standalone bots like Google's Gemini and OpenAI's ChatGPT that are powered by their own large language models, DuckDuckGo's AI Chat is not. Instead, think of it as a way to access multiple chatbots in a single place. Right now, AI chat will let you choose between OpenAI's GPT 3.5, Anthropic's Claude 3 Haiku, Meta's Llama 3 and Mistral's Mistral 8x7B, and the company says that more models are coming soon.

Until mid-2023, artificial intelligence was something of a niche topic in Washington, largely confined to small circles of tech-policy wonks. That all changed when, nearly two years into Gina Raimondo's tenure as Secretary of Commerce, ChatGPT's explosive popularity catapulted AI into the spotlight. Raimondo, however, was ahead of the curve. "I make it my business to stay on top of all of this," she says during an interview in her wood-paneled office overlooking the National Mall on May 21. "None of it was shocking to me." But in the year since, even she has been startled by the pace of progress.

Microsoft, OpenAI and Nvidia face increased antitrust scrutiny of their roles in the artificial intelligence industry after a report that US regulators have reached an agreement on investigating the companies. The New York Times reported that the US justice department and the Federal Trade Commission (FTC) have reached an agreement on investigations into the main protagonists in the AI market. The deal is expected to be completed in the coming days, according to the report. The justice department will lead on investigating whether Nvidia, the leading maker of chips that train and operate AI systems, has broken antitrust laws that oversee fair competition in business and aim to prevent monopolies, said the NYT on Wednesday. The Wall Street Journal also reported on Thursday that the FTC is investigating whether Microsoft structured a recent deal with startup Inflection AI to avoid an antitrust inquiry.

We introduce Buffer of Thoughts (BoT), a novel and versatile thought-augmented reasoning approach for enhancing accuracy, efficiency and robustness of large language models (LLMs). Specifically, we propose meta-buffer to store a series of informative high-level thoughts, namely thought-template, distilled from the problem-solving processes across various tasks. Then for each problem, we retrieve a relevant thought-template and adaptively instantiate it with specific reasoning structures to conduct efficient reasoning. To guarantee the scalability and stability, we further propose buffer-manager to dynamically update the meta-buffer, thus enhancing the capacity of meta-buffer as more tasks are solved. We conduct extensive experiments on 10 challenging reasoning-intensive tasks, and achieve significant performance improvements over previous SOTA methods: 11% on Game of 24, 20% on Geometric Shapes and 51% on Checkmate-in-One. Further analysis demonstrate the superior generalization ability and model robustness of our BoT, while requiring only 12% of the cost of multi-query prompting methods (e.g., tree/graph of thoughts) on average. Notably, we find that our Llama3-8B + BoT has the potential to surpass Llama3-70B model.

One type of question that is commonly found in day-to-day scenarios is ``fan-out'' questions, complex multi-hop, multi-document reasoning questions that require finding information about a large number of entities. However, there exist few resources to evaluate this type of question-answering capability among large language models. To evaluate complex reasoning in LLMs more fully, we present FanOutQA, a high-quality dataset of fan-out question-answer pairs and human-annotated decompositions with English Wikipedia as the knowledge base. We formulate three benchmark settings across our dataset and benchmark 7 LLMs, including GPT-4, LLaMA 2, Claude-2.1, and Mixtral-8x7B, finding that contemporary models still have room to improve reasoning over inter-document dependencies in a long context. We provide our dataset and open-source tools to run models to encourage evaluation at https://fanoutqa.com

In-Context Learning (ICL) has been a powerful emergent property of large language models that has attracted increasing attention in recent years. In contrast to regular gradient-based learning, ICL is highly interpretable and does not require parameter updates. In this paper, we show that, for linearized transformer networks, ICL can be made explicit and permanent through the inclusion of bias terms. We mathematically demonstrate the equivalence between a model with ICL demonstration prompts and the same model with the additional bias terms. Our algorithm (ICLCA) allows for exact conversion in an inexpensive manner. Existing methods are not exact and require expensive parameter updates. We demonstrate the efficacy of our approach through experiments that show the exact incorporation of ICL tokens into a linear transformer. We further suggest how our method can be adapted to achieve cheap approximate conversion of ICL tokens, even in regular transformer networks that are not linearized. Our experiments on GPT-2 show that, even though the conversion is only approximate, the model still gains valuable context from the included bias terms.

In-context learning (ICL) empowers large language models (LLMs) to tackle new tasks by using a series of training instances as prompts. Since generating the prompts needs to sample from a vast pool of instances and annotate them (e.g., add labels in classification task), existing methods have proposed to select a subset of unlabeled examples for annotation, thus enhancing the quality of prompts and concurrently mitigating annotation costs. However, these methods often require a long time to select instances due to their complexity, hindering their practical viability. To address this limitation, we propose a graph-based selection method, FastGAS, designed to efficiently identify high-quality instances while minimizing computational overhead. Initially, we construct a data similarity graph based on instance similarities. Subsequently, employing a graph partitioning algorithm, we partition the graph into pieces. Within each piece (i.e., subgraph), we adopt a greedy approach to pick the most representative nodes. By aggregating nodes from diverse pieces and annotating the corresponding instances, we identify a set of diverse and representative instances for ICL. Compared to prior approaches, our method not only exhibits superior performance on different tasks but also significantly reduces selection time. In addition, we demonstrate the efficacy of our approach in LLMs of larger sizes.