Multimodal large language models (MLLMs) have shown promising capabilities but struggle under distribution shifts, where evaluation data differ from instruction tuning distributions. Although previous works have provided empirical evaluations, we argue that establishing a formal framework that can characterize and quantify the risk of MLLMs is necessary to ensure the safe and reliable application of MLLMs in the real world. By taking an information-theoretic perspective, we propose the first theoretical framework that enables the quantification of the maximum risk of MLLMs under distribution shifts. Central to our framework is the introduction of Effective Mutual Information (EMI), a principled metric that quantifies the relevance between input queries and model responses. We derive an upper bound for the EMI difference between in-distribution (ID) and out-of-distribution (OOD) data, connecting it to visual and textual distributional discrepancies. Extensive experiments on real benchmark datasets, spanning 61 shift scenarios empirically validate our theoretical insights.

Clustering is at the very core of machine learning, and its applications proliferate with the increasing availability of data. However, as datasets grow, comparing clusterings with an adjustment for chance becomes computationally difficult, preventing unbiased ground-truth comparisons and solution selection. We propose FastAMI, a Monte Carlo-based method to efficiently approximate the Adjusted Mutual Information (AMI) and extend it to the Standardized Mutual Information (SMI). The approach is compared with the exact calculation and a recently developed variant of the AMI based on pairwise permutations, using both synthetic and real data. In contrast to the exact calculation our method is fast enough to enable these adjusted information-theoretic comparisons for large datasets while maintaining considerably more accurate results than the pairwise approach.

Applying for a frontline job can be a game of hurry-up-and-wait, and communication is not always the best when a company is trying to fill dozens of positions at the same time. Enter Emi, the latest company targeting technology to this portion of the workforce with a conversational artificial intelligence recruiting tool. The technology automates communication between global enterprises and candidates using a conversational interface. CEO Mateo Cavasotto says this reduces the time it takes to hire people, while also increasing candidate satisfaction, thus improving recruitment productivity. The idea for the company came a couple of years ago when Cavasotto and Andres Arslanian, CTO, worked as volunteers for a Microcredits NGO in Argentina. They were working to understand how problems among the poverty-stricken population could be solved with technology.

In 1950, Alan Turing developed the Turing Test as a test of a machine's ability to display human-like intelligent behavior. "Are there imaginable digital computers which would do well in the imitation game?" In most applications of AI, a model is created to imitate the judgment of humans and implement it at scale, be it autonomous vehicles, text summarization, image recognition, or product recommendation. By the nature of imitation, a computer is only able to replicate what humans have done, based on previous data. This doesn't leave room for genuine creativity, which relies on innovation, not imitation.

Reinforcement learning could be hard when the reward signal is sparse. In these scenarios, exploration strategy becomes essentially important: a good exploration strategy not only helps the agent to gain a faster and better understanding of the world but also makes it robust to the change of the environment. In this article, we discuss a novel exploration method, namely Exploration with Mutual Information(EMI) proposed by Kim et al. in ICML 2019. In a nutshell, EMI learns representations for both observations(states) and actions in the expectation that we can have a linear dynamics model on these representations. EMI then computes the intrinsic reward as the prediction error under the linear dynamics model.

The first testing sessions for SampleRNN – an artificially intelligent software developed by computer scientist duo CJ Carr and Zach Zukowski, AKA Dadabots – sounded more like a screamo gig than a machine-learning experiment. Carr and Zukowski hoped their program could generate full-length black metal and math rock albums by feeding it small chunks of sound. The first trial consisted of encoding and entering in a few Nirvana a cappellas. "When it produced its first output," Carr tells me over email, "I was expecting to hear silence or noise because of an error we made, or else some semblance of singing. The first thing it did was scream about Jesus. We looked at each other like, 'What the fuck?'"

Policy optimization struggles when the reward feedback signal is very sparse and essentially becomes a random search algorithm until the agent accidentally stumbles upon a rewarding or the goal state. Recent works utilize intrinsic motivation to guide the exploration via generative models, predictive forward models, or more ad-hoc measures of surprise. We propose EMI, which is an exploration method that constructs embedding representation of states and actions that does not rely on generative decoding of the full observation but extracts predictive signals that can be used to guide exploration based on forward prediction in the representation space. Our experiments show the state of the art performance on challenging locomotion task with continuous control and on image-based exploration tasks with discrete actions on Atari.

The possibilities clog our news feeds, create interesting conversations, and give tech leaders inspiration to explore solutions. What will this mean for us as humans? Could this impact all of society? With all the questions being asked, only one thing is absolutely clear. We're about to enter one of the biggest transformations our society has witnessed in the last century - if not millennium.

It was written not by a human being, but by my computer program EWI (an acronym for "experiments in writing intelligence"). EWI was fed the texts of two of Hofstadter's books--namely, Gödel, Escher, Bach (winner of the Pulitzer Prize for General Nonfiction in 1980) and Metamagical Themas--and then, following its code, EWI carefully analyzed these two books for their uniquely Hofstadterian stylistic elements and features, after which it recombined these stylistic elements in new fashions. EWI thereby came up with some 25 new and highly diverse "Hofstadter articles," one of which is given below, and the article is followed by a brief commentary about EWI and its output by Hofstadter himself. Actually, I should state up front that the wonderful sparkling dialogues of GEB, which are a substantial part of that book, were not used by EWI in generating any of the articles, because EWI is unfortunately not yet able to work with inputs that belong to different genres, such as chapters and dialogues. To combine stylistic aspects of two or more different genres of writing represents a very thorny challenge indeed.

It's become customary to be wowed by computers today. IBM's supercomputer, Watson, beat former Jeopardy! More recently, Google's AlphaGo eclipsed Lee Sedol in a challenge of Go, an ancient board game. These computer systems used artificial intelligence (AI) to outshine humans. While board and game show victories are exciting, they overshadow the practical benefits of AI.