Neural Information Processing Systems http://nips.cc/

A week after its algorithms advised people to eat rocks and put glue on pizza, Google admitted Thursday that it needed to make adjustments to its bold new generative AI search feature. The episode highlights the risks of Google's aggressive drive to commercialize generative AI--and also the treacherous and fundamental limitations of that technology. Google's AI Overviews feature draws on Gemini, a large language model like the one behind OpenAI's ChatGPT, to generate written answers to some search queries by summarizing information found online. The current AI boom is built around LLMs' impressive fluency with text, but the software can also use that facility to put a convincing gloss on untruths or errors. Using the technology to summarize online information promises can make search results easier to digest, but it is hazardous when online sources are contractionary or when people may use the information to make important decisions.

In an unassuming office on a quiet, mostly residential street in Mountain View, California -- located eight minutes from Google's sprawling headquarters -- a couple of ex-Googlers and their team of 50 are trying to build a search engine they hope will someday rival their former employer's. The company, Neeva, was started in 2020 by Sridhar Ramaswamy, who ran Google's $162 billion advertising arm before stepping down in 2018, and Vivek Raghunathan, a former Google vice president who worked on monetizing YouTube and other parts of the company. For a few years, the startup, which has raised over $77 million from some of Silicon Valley's top investors, focused on differentiating itself from Google by shunning invasive advertising and allowing power users to pay for extra features. Then, around the end of last year, the team at Neeva watched as a chatbot called ChatGPT created by the San Francisco–based startup OpenAI went viral. ChatGPT's ability to divine answers to nearly every question with an eerily humanlike sentience made it an instant hit, unleashing a modern AI wave. Suddenly, people around the world were talking about replacing Google search with ChatGPT. After all, if a chatbot could instantly answer any question for you, why would you need a search engine that simply spat out a bunch of links for you to trawl through?

You.com founder Richard Socher knows that his company has always been a David going after the Goliath in search, Google, and to a lesser extent Microsoft. He likes to point out that his company built search based on generative AI in December, several months before the other giant search players made their announcements. Today, the company is announcing it's taking that head start and building on it with multimodal search. That means it can add elements beyond text to help answer a question more precisely. So say you ask a question such as "Which company has the most CRM market share," you will get the answer "Salesforce," and if you follow up with "What is Saleforce's stock price?",

In recent years, Multi-Agent Path Finding (MAPF) has attracted attention from the fields of both Operations Research (OR) and Reinforcement Learning (RL). However, in the 2021 Flatland3 Challenge, a competition on MAPF, the best RL method scored only 27.9, far less than the best OR method. This paper proposes a new RL solution to Flatland3 Challenge, which scores 125.3, several times higher than the best RL solution before. We creatively apply a novel network architecture, TreeLSTM, to MAPF in our solution. Together with several other RL techniques, including reward shaping, multiple-phase training, and centralized control, our solution is comparable to the top 2-3 OR methods.

At least, that's the crux of the argument Richard Socher, the former chief scientist at Salesforce, likes to make. In 2020, Socher co-founded You, a search engine that uses AI to understand search queries, rank the results and parse the queries into different languages (including programming languages). You summarizes information from across the web and offers built-in apps, like search tools for Twitter, that allow users to complete tasks without having to leave the results page. It seems there's some truth to his words. Socher claims that You has hundreds of thousands of users, with 70% growth in sign-ups last month and 30% growth in unique searches month over month.

Learning against label noise is a vital topic to guarantee a reliable performance for deep neural networks. Recent research usually refers to dynamic noise modeling with model output probabilities and loss values, and then separates clean and noisy samples. These methods have gained notable success. However, unlike cherry-picked data, existing approaches often cannot perform well when facing imbalanced datasets, a common scenario in the real world. We thoroughly investigate this phenomenon and point out two major issues that hinder the performance, i.e., \emph{inter-class loss distribution discrepancy} and \emph{misleading predictions due to uncertainty}. The first issue is that existing methods often perform class-agnostic noise modeling. However, loss distributions show a significant discrepancy among classes under class imbalance, and class-agnostic noise modeling can easily get confused with noisy samples and samples in minority classes. The second issue refers to that models may output misleading predictions due to epistemic uncertainty and aleatoric uncertainty, thus existing methods that rely solely on the output probabilities may fail to distinguish confident samples. Inspired by our observations, we propose an Uncertainty-aware Label Correction framework~(ULC) to handle label noise on imbalanced datasets. First, we perform epistemic uncertainty-aware class-specific noise modeling to identify trustworthy clean samples and refine/discard highly confident true/corrupted labels. Then, we introduce aleatoric uncertainty in the subsequent learning process to prevent noise accumulation in the label noise modeling process. We conduct experiments on several synthetic and real-world datasets. The results demonstrate the effectiveness of the proposed method, especially on imbalanced datasets.

But it falls short of the definition of artificial general intelligence (AGI), which would be a machine capable of understanding the world as well as any human. In the 1950s, researchers including AI pioneer Herbert A. Simon were convinced that AGI would exist within the next few decades. Since then, AGI has proven to be a daunting, perhaps even impossible-to-achieve milestone. Writing in The Guardian, roboticist Alan Winfield claimed the gulf between modern computing and AGI is as wide as the gulf between current space flight and faster-than-light travel. Still, others insist that AGI is drawing close within reach.

Salesforce is shutting down two of its AI-powered voice services -- Einstein Voice Assistant and Voice Skills -- as it shifts resources toward its newly released Salesforce Anywhere app, as spotted by Voicebot.ai. A company spokesperson told VentureBeat that voice capabilities remain "a priority" for Salesforce, and that the products it's discontinuing will inform the development of "reimagined" functionality focused on productivity and collaboration. Einstein Voice Assistant, which launched in beta last year, was a component of Salesforce's Einstein Voice -- an outgrowth of the company's Einstein technology that enables customers to navigate cloud services hands-free. One of its ostensible advantages over other platforms was its versatility: It was siloed, restricting data pulls to individual users' accounts, and it could be "taught" to recognize jargon, acronyms, and slang in an organization's lexicon. Einstein Voice Assistant was more than a glorified transcriber.

Tax policy is surely a complex beast, and depending on your political leanings, you probably have some strong feelings about how it should be implemented. Salesforce AI researchers are trying to build a model to bring artificial intelligence to bear on what will undoubtedly always be a highly political process. Richard Socher, who heads up AI research at Salesforce, says the company is researching all kinds of solutions related to AI and business, and how it could improve the Salesforce product family; however, he also looks at how his team could use AI to solve a set of broader social issues beyond what it can do for the product line. Socher says when you look at the biggest issues of our time, one of the largest is economic inequality, and how we could use policy to solve that. To that end, the company created a model it calls an AI economist that could look at various economic variables, a broad set of economic models and using the power of AI begin to demonstrate how various policies affect economic inequality versus productivity.