Google has made significant progress on artificial intelligence (AI) reasoning software across multiple teams in the past few months. These advancements focus on tackling multi-step problems in fields such as mathematics and computer programming.
Google is employing a technique called “chain-of-thought prompting” to simulate human reasoning, similar to OpenAI's approaches. This innovation aims to help Google maintain its competitive edge in the rapidly evolving AI sector. Alongside, Google is also developing an AI lab assistant to improve research efficiency, marking a new frontier in the competition between Google and OpenAI.
Some Google employees expressed concern over potentially lagging behind OpenAI, especially after OpenAI's release of the GPT-4 model, which stirred fears among investors about diminishing demand for Google's search engine. In response, Google has merged its top research labs into the Google DeepMind division and strengthened collaboration between researchers and product teams.
Despite worries, recent demonstrations of Google's AI projects have alleviated some concerns. Analysts believe Google's tech capabilities remain top-tier, although the company has been more conservative in product launches, taking into account ethical issues and maintaining brand trust.
Furthermore, Google DeepMind, under the leadership of Sir Demis Hassabis, is working on an AI lab assistant designed to help researchers plan and predict the outcomes of scientific experiments. This tool aims to foster interdisciplinary collaboration and uncover unexpected connections in research, representing a potential revolution in biology.
Other companies are also pursuing specialized applications for data-intensive AI models. For instance, Germany's BioNTech and its AI subsidiary InstaDeep have developed Laila, an AI assistant with detailed biological knowledge based on Meta's open-source Llama 3.1 model. Laila automates routine tasks like DNA sequencing and experiment visualization, increasing productivity for scientists and technicians.
According to InstaDeep's CEO, Karim Beguir, AI agents like Laila are seen as productivity tools rather than fully automated solutions, allowing scientists to focus on more critical tasks.
