We propose Re-FORC, an adaptive reward prediction method that, given a context, enables prediction of the expected future rewards as a function of the number of future thinking tokens. Re-FORC trains a lightweight adapter on reasoning models, demonstrating improved prediction with longer reasoning and larger models. Re-FORC enables: 1) early stopping of unpromising reasoning chains, reducing compute by 26% while maintaining accuracy, 2) optimized model and thinking length selection that achieves 4% higher accuracy at equal compute and 55% less compute at equal accuracy compared to the largest model, 3) adaptive test-time scaling, which increases accuracy by 11% in high compute regime, and 7% in low compute regime. Re-FORC allows dynamic reasoning with length control via cost-per-token thresholds while estimating computation time upfront.

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 - 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Robotics is a diverse industry with many variables. Its future is filled with uncertainty: nobody can predict which way it will develop and what directions will be leading a few years from now. According to the International Federation of Robotics data, 3 million industrial robots are operating worldwide – the number has increased by 10% over 2021. The global robotics market is estimated at $55.8 billion and is expected to grow to $91.8 billion by 2026 with a 10.5% annual growth rate.

More than two dozen artificial intelligence experts from business and academia, including Texas McCombs, explored the importance of understanding how machine learning systems arrive at their conclusions so humans can trust those results. This relatively new frontier of explainable AI, or XAI, was scrutinized for two half-day sessions in November during the online Global Analytics Summit held by the McCombs Center for Analytics and Transformative Technologies (CATT). More than 3,500 people registered for the conference, another step in the university's drive to be a thought leader in the field. Experts say that as AI systems become more sophisticated, they become increasingly difficult for humans to interpret. The calculation process turns into what is called a "black box" that even data scientists who create the algorithms can't understand, according to IBM Corp. XAI methods, however, provide transparency so humans can understand the results and assess their accuracy and fairness.

Goal-conditioned reinforcement learning (GCRL), related to a set of complex RL problems, trains an agent to achieve different goals under particular scenarios. Compared to the standard RL solutions that learn a policy solely depending on the states or observations, GCRL additionally requires the agent to make decisions according to different goals. In this survey, we provide a comprehensive overview of the challenges and algorithms for GCRL. Firstly, we answer what the basic problems are studied in this field. Then, we explain how goals are represented and present how existing solutions are designed from different points of view. Finally, we make the conclusion and discuss potential future prospects that recent researches focus on.

The creative spark is one of humanity's defining features. But increasingly powerful artificial intelligence (AI) is bringing innovation and automation together to create new tools for invention. Eureka moments tend to come from the rare convergence of ideas from diverse fields, says Julian Nolan, chief executive of Iprova. However, growing specialisation among experts means this is becoming even rarer, he adds, often relying on chance meetings at conferences or a fortuitous conversation at the office coffee machine. "How crazy is it that something so key to corporations is based on human serendipity?" says Mr Nolan.

Artificial Intelligence (AI) has long pursued models, theories, and techniques to imbue machines with human-like general intelligence. Yet even the currently predominant data-driven approaches in AI seem to be lacking humans' unique ability to solve wide ranges of problems. This situation begs the question of the existence of principles that underlie general problem-solving capabilities. We approach this question through the mathematical formulation of analogies across different problems and solutions. We focus in particular on problems that could be represented as tree-like structures. Most importantly, we adopt a category-theoretic approach in formalising tree problems as categories, and in proving the existence of equivalences across apparently unrelated problem domains. We prove the existence of a functor between the category of tree problems and the category of solutions. We also provide a weaker version of the functor by quantifying equivalences of problem categories using a metric on tree problems.

AI is buzzing in the legal industry. It's time to separate hype from reality. I was recently invited to speak on artificial intelligence in law at the annual LMA conference along with an outstanding panel of industry experts (read my summary here). During my talk I offered a simplified view of the AI landscape of vendors and technologies in play and received great feedback and many requests for further detail. This feedback is consistent with many conversations I've had regarding legal AI over the past few years: it's hard to distinguish fact from reality, and practical examples of AI in this industry are hard to come by.

Speedup learning seeks to improve the computational efficiency of problem solving with experience. In this paper, we develop a formal framework for learning efficient problem solving from random problems and their solutions. We apply this framework to two different representations of learned knowledge, namely control rules and macro-operators, and prove theorems that identify sufficient conditions for learning in each representation. Our proofs are constructive in that they are accompanied with learning algorithms. Our framework captures both empirical and explanation-based speedup learning in a unified fashion. We illustrate our framework with implementations in two domains: symbolic integration and Eight Puzzle. This work integrates many strands of experimental and theoretical work in machine learning, including empirical learning of control rules, macro-operator learning, Explanation-Based Learning (EBL), and Probably Approximately Correct (PAC) Learning.