Skip to content

Google unveils PaLM 2, its new AI LLM


Introduction: Google I/O Conference and AI Focus

Google kicked off its annual I/O conference today with a core focus on advancing artificial intelligence (AI) across its domain. The conference this year is centered around AI, as Google aims to establish a leadership position in the market. This comes as Microsoft and OpenAI enjoy the success of ChatGPT.

PaLM 2: Google’s Powerful Language Model

The foundation of Google’s AI efforts lies in its new PaLM 2 large language model (LLM). PaLM 2 is set to power at least 25 Google products and services, including Bard, Workspace, Cloud, Security, and Vertex AI. It is an enhanced version of the original PaLM, expanding Google’s generative AI capabilities significantly.

Google’s Vision for AI

During a press briefing, Zoubin Ghahramani, VP of Google DeepMind, emphasized the mission of making information universally accessible and useful. Ghahramani highlighted that AI is enabling a deeper understanding of the world and enhancing the helpfulness of Google’s products.

PaLM 2 Capabilities and Improvements

Ghahramani described PaLM 2 as a state-of-the-art language model proficient in math, coding, reasoning, multilingual translation, and natural language generation. The model surpasses its predecessors in various ways, although Ghahramani did not disclose the specific parameter size, as it does not solely determine performance or capability.

PaLM 2 was trained using Google’s latest Tensor Processing Unit (TPU) infrastructure, optimizing its training process. The model also demonstrates enhanced AI inference through improved compute, scaling, dataset mixtures, and model architectures.

Improved Core Capabilities of PaLM 2

  • Multilinguality: PaLM 2 is trained on over 100 spoken-word languages, enabling it to excel in multilingual tasks and understand nuanced phrases and figurative language.
  • Reasoning: The model offers stronger logic, common sense reasoning, and mathematical capabilities, having been trained on a vast amount of math and science texts, including scientific papers and mathematical expressions.
  • Coding: PaLM 2 comprehends, generates, and debugs code, being pretrained on over 20 programming languages, including Python, JavaScript, and older languages like Fortran. It can even provide documentation and support programmers worldwide.

PaLM 2 Applications and Integration

Ghahramani highlighted that PaLM 2 is adaptable to a wide range of tasks and is being integrated into various Google products. The Med-PaLM 2 model caters to the medical profession, while Sec-PaLM focuses on security use cases. Bard, Google’s competitor to ChatGPT, will benefit from PaLM 2’s power, providing an intuitive prompt-based user interface. Google Workspace applications will also receive an intelligence boost with PaLM 2.


Google’s focus on AI at the I/O conference showcases its commitment to advancing technology. PaLM 2, with its improved capabilities and integration across multiple products and services, aims to enhance user experiences and enable developers to leverage powerful language models for various tasks.


What is PaLM 2?

PaLM 2 is a large language model developed by Google to power numerous products and services. It excels in math, coding, reasoning, multilingual translation, and natural language generation.

How does PaLM 2 differ from previous language models?

PaLM 2 surpasses previous models in terms of performance and capabilities. It has been trained using Google’s advanced TPU infrastructure, making it more efficient for AI inference. The model’s improved core capabilities include multilinguality, stronger reasoning, and coding proficiency.

Which Google products benefit from PaLM 2?

PaLM 2 powers at least 25 Google products, including Bard, Workspace, Cloud, Security, and Vertex AI. It is a versatile model that can be fine-tuned for specific tasks, enhancing user experiences across various domains.


Source Link