Google's Blueshift Team Joins DeepMind to Catch Up with ChatGPT's Progress in Large Language Models

Google’s Blueshift Team, which specializes in large language models, has announced its merger with DeepMind, another major player in the artificial intelligence (AI) industry. The team’s aim is to improve the capabilities of large language models, and the move is seen as a response to the recent rise of ChatGPT. The Blueshift team is made up of four members who have contributed to the development of Minerva, a mathematical problem-solving model that has performed exceptionally well in exams such as MATH and MMLU-STEM. Additionally, the team has worked on training large models to understand code debugging and provided advice for Google’s 540 billion large model PaLM. This article examines the impact of the Blueshift team’s merger with DeepMind, Google’s response to ChatGPT, and its future plans for large models.

As the demand for advanced AI models continues to rise, major tech companies such as Google are investing heavily in large language models (LLMs). The Blueshift team, affiliated with Google Research, has announced its merger with DeepMind to improve LLM capabilities. This move is seen as a response to the recent rise of ChatGPT and aims to help Google catch up with its competitors in the AI industry.

Who is the Blueshift team?

The Blueshift team is made up of four members who have contributed significantly to the development of Minerva, a mathematical problem-solving model. The team's research focuses on understanding the limitations of the Transformer and challenging its extension to address areas such as mathematics, science, programming, algorithms, and planning. Their work on Minerva has enabled the model to understand mathematical symbols in the same way as natural language after reading papers and LaTeX formulas. The model has performed exceptionally well in exams such as MATH and MMLU-STEM. Additionally, the team has worked on training large models to understand code debugging and provided advice for Google’s 540 billion large model PaLM.

Google’s response to ChatGPT

ChatGPT has been a significant disruptor in the AI industry, and Google has been working to catch up with its competitors. Google Research launched the BIG-Bench, a new benchmark for large models in 2021, which has since been contributed to by the Blueshift team. Additionally, Google released its 540 billion large model PaLM, which the Blueshift team also provided advice for. However, despite these efforts, Google has been criticized for the poor performance of its latest LLM, Bard.

Impact of Blueshift Team’s merger with DeepMind

The Blueshift team’s merger with DeepMind is seen as a response to the recent rise of ChatGPT and aims to accelerate the improvement of LLM capabilities. The move is expected to improve Google's competitiveness in the AI industry and help it catch up with its competitors. DeepMind has already developed its dialogue AI, Sparrow, which is based on reinforcement learning based on human feedback and can train models based on human preferences. The beta version of Sparrow is expected to be released in 2023.

Google’s future plans for large models

Google has announced its plans to spend $300 million on Anthropic, a company founded by the core members of GPT-3, a competing ChatGPT company. This investment is expected to help Google catch up with its competitors and further improve its large model capabilities. In addition, Google is expected to release the internal test version of Sparrow this year, which may benefit from the expertise of the Blueshift team.