On Thursdays, we dive deep into one of the freshest research papers or technology frameworks that is worth your attention. Use the massive 50% OFF and become a Premium member. ONLY 5 DAYS LEFT →

SUBSCRIBE TODAY FOR $25/per year

💥 What’s New in AI: Google’s Minerva Can Solve Mathematical and Scientific Problems Through Step-by-Step Reasoning

Quantitative reasoning (QR) is one of the crown jewels of the new generation of deep learning methods. While recently we have seen major breakthroughs in natural language understanding (NLU) techniques, the question of whether this type of model can exhibit reasoning capabilities remains unanswered. From all the reasoning benchmarks in language models, nothing ranks as high as solving mathematical or scientific textual problems. A few months ago, OpenAI astonished the world with the release of Lean, a model that was able to solve problems from the math Olympiads. A few days ago, Google Research published a paper discussing Minerva, a model that was able to solve mathematical or scientific problems using step-by-step reasoning.

Solving mathematical problems in textual forms requires a combination of skills such as interpreting the textual input, recalling relevant math formulas and correct symbolic manipulations to generate the step-by-step processes. Not surprisingly, quantitative reasoning requires a combination of different deep learning techniques that are often non-trivial to make them work together. With Minerva…→let’s dive deeper