Google is developing AI reasoning software similar to OpenAI's GPT-3, AI Lab Assistant

Media reports that Google is developing an artificial intelligence software that mimics human reasoning abilities, similar to OpenAI's o1 model. Analysis suggests that this marks a new frontier of competition between Google and the rapidly growing OpenAI. At the same time, Google is also developing an AI laboratory assistant product to help laboratories improve research efficiency.

Concerns about falling too far behind OpenAI but still competitive

According to sources familiar with the matter, in recent months, multiple teams at Google have made progress in artificial intelligence reasoning software, which performs better in solving multi-step problems in fields such as mathematics and computer programming.

Currently, Google's artificial intelligence researchers are trying to drive the next major advancement in technology. Like OpenAI, Google is also attempting to use a technique called "chain-of-thought prompting" to simulate human reasoning. According to two sources, this technique was originally developed by Google, where the software pauses for a few seconds before responding to written prompts, considers a series of related prompts invisibly to the user, and then summarizes what appears to be the optimal response.

Since OpenAI introduced its o1 model, Google has been competing with OpenAI for dominance in the field of artificial intelligence, especially after OpenAI launched the popular chatbot ChatGPT, which has raised concerns among some investors that the demand for Google search may gradually diminish.

To regain its leading position, Google has taken various measures, including merging its top research labs into the Google DeepMind department and strengthening the relationship between researchers and product teams. However, Google is still relatively slow in launching artificial intelligence products, mainly because it needs to consider ethical issues, maintain public trust in its brand, and balance the competitive interests of multiple similar projects within the company.

According to another source, since OpenAI introduced its o1 model in mid-September, some employees at Google DeepMind have been concerned that the company is falling behind. However, as Google has showcased its work, these concerns have eased.

Analysis suggests that despite Google's slower product release pace, it remains a strong competitor. Oren Etzioni, a senior artificial intelligence researcher and founder of TrueMedia.org, stated:

"Technically, Google's capabilities have always been top-notch, they are just more conservative in product releases. This is a marathon, and anyone has a chance to win the race."

Developing an AI Laboratory Assistant

Furthermore, according to the Financial Times, Google DeepMind is developing an AI laboratory assistant to help researchers plan scientific experiments and better predict experimental results. Led by Google's AI department head Sir Demis Hassabis, the project aims to help scientists collaborate across disciplines and more easily discover unexpected connections.

At a recent event hosted by the Nobel Foundation, Hassabis stated that artificial intelligence software is sparking a "revolution" in the field of biology

"We are developing a large scientific language model that can act as a research assistant and may even help you predict the results of experiments."

He also stated that in the coming years, tools being developed by DeepMind can suggest and design experiments based on given hypotheses, providing scientists with a probabilistic view of the likelihood of experiment success or failure.

Currently, companies are actively seeking specialized applications for energy- and data-intensive artificial intelligence models. German pharmaceutical company BioNTech and its London-based AI subsidiary InstaDeep have designed a specialized AI assistant named Laila, equipped with "detailed biological knowledge" and built on Meta's open-source Llama 3.1 model.

In a live demonstration, research scientist Arnu Pretorius showcased how the AI agent automates routine tasks in experimental biology, such as analyzing and segmenting DNA sequences, as well as visualizing experimental results.

BioNTech scientists also demonstrated how Laila connects to laboratory equipment, monitors ongoing experiments, or tasks carried out by robots. During the live demonstration, the assistant even detected a mechanical failure in a BioNTech machine.

"We do not believe that the future will be completely dominated by AI automation. We see AI agents like Laila as tools to enhance productivity, enabling scientists and technicians to focus their limited time on truly important matters," said Karim Beguir, CEO of InstaDeep