Large Language Models still struggle in challenging scenarios that leverage structured data, complex reasoning, or tool usage. In this paper, we propose Source2Synth: a new method that can be used for teaching LLMs new skills without relying on costly human annotations. Source2Synth takes as input a custom data source and produces synthetic data points with intermediate reasoning steps grounded in real-world sources. Source2Synth improves the dataset quality by discarding low-quality generations based on their answerability. We demonstrate the generality of this approach by applying it to two challenging domains: we test reasoning abilities in multi-hop question answering (MHQA), and tool usage in tabular question answering (TQA). Our method improves performance by 25.51% for TQA on WikiSQL and 22.57% for MHQA on HotPotQA compared to the fine-tuned baselines.

Adobe is leaning on AI-powered data solutions to bridge the post-cookie identity gap. This fits into Adobe's broader strategy of using a mixture of automation and artificial intelligence to figure out what people are looking for and predict how brands can demonstrate value for customers in the moments that matter, said Kevin Lindsay, Adobe's director of product marketing. In practice, that means focusing on reducing churn and anticipating a customer's needs rather than just pushing to complete a transaction. Considering the rising cost of customer acquisition, convincing someone not to cancel a service can be more valuable than converting a new customer. "It's also about paying attention to signals and emotional cues, like frustration," Lindsay said, and [determining whether you're] ticking people off with a bad experience."

Calorie and nutrition research has attained increased interest in recent years. But, due to the complexity of the problem, literature in this area focuses on a limited subset of ingredients or dish types and simple convolutional neural networks or traditional machine learning. Simultaneously, estimation of ingredient portions can help improve calorie estimation and meal re-production from a given image. In this paper, given a single cooking image, a pipeline for calorie estimation and meal re-production for different servings of the meal is proposed. The pipeline contains two stages. In the first stage, a set of ingredients associated with the meal in the given image are predicted. In the second stage, given image features and ingredients, portions of the ingredients and finally the total meal calorie are simultaneously estimated using a deep transformer-based model. Portion estimation introduced in the model helps improve calorie estimation and is also beneficial for meal re-production in different serving sizes. To demonstrate the benefits of the pipeline, the model can be used for meal kits generation. To evaluate the pipeline, the large scale dataset Recipe1M is used. Prior to experiments, the Recipe1M dataset is parsed and explicitly annotated with portions of ingredients. Experiments show that using ingredients and their portions significantly improves calorie estimation. Also, a visual interface is created in which a user can interact with the pipeline to reach accurate calorie estimations and generate a meal kit for cooking purposes.

Having been privileged to witness the evolution of the data science and artificial intelligence (AI) scene in the Middle East for the past 10 years and having spoken at one of the first big data events in Dubai back in 2013, it is clear to me that there are considerable opportunities for AI in this vibrant region. Recently, I got the chance to present on the top 10 AI challenges of companies in the Gulf Cooperation Council (GCC) region, at Virtual Executive Boardroom: Key Insights on Becoming a Data-Driven Enterprise, which took place at DigiConnect (UAE) and was delivered to top C-level executives and senior data managers from the most relevant companies in the GCC region. In this post, I will not get into each of these ten challenges. However, I will focus on a common issue that came up as a top priority for them in a quick live poll during the session: The issue of trusting decisions made by AI. Interpreting deep learning networks takes place in a tough playground, so making AI interpretable serves one specific goal, and that is to trust the decisions made by AI models.

PARIS – MDMA, the main ingredient in ecstasy, makes humans more likely to cooperate -- but only with trustworthy people -- researchers said Monday in the first study into how the drug impacts our willingness to help others. Despite its status in Britain as a Class A drug, MDMA is widely consumed due to the heightened sense of energy, empathy and pleasure it arouses in users. It contains neurotransmitters -- chemical messengers for the brain -- that are known to be linked to behavior and mood, but scientists currently understand very little about how these affect social interactions. Researchers at King's College London studied 20 healthy adult men who were given a typical recreational dose of MDMA or a placebo pill and then asked to complete a set of tasks while images of their brain activity were taken with an MRI scanner. One of the mind exercises they were given was the Prisoner's Dilemma -- an example of so-called game theory in which an individual is asked to choose between cooperating or competing with another, unknown person.