Image Source: Google
Google unveiled PaLM 2, its latest large language model (LLM) at Google I/O 2023 alongside much-anticipated devices like the Pixel Fold, Pixel Tablet and Pixel 7a. This powerful model will be utilised to update Google’s crown jewel, Search, its chatbot Bard — OpenAI’s ChatGPT rival — and Workspace.
The PaLM 2 is built on top of Goggle’s JAX and TPU v4 infrastructure. Google is building several models on top of their latest LLM and a few of them are out for preview. These include four different sizes of the system, which are called Gecko, Otter, Bison and Unicorn, arranged from smallest to largest. These variants have been customised using domain-specific data to execute specific functions for enterprise customers.
PaLM 2 is described by Google as a range of models which includes Codey and Med-PaLM 2, which is a medical knowledge-oriented model. Additionally, there is Sec-PaLM, which concentrates on security use cases and a smaller version of PaLM 2 designed to run on smartphones. This could potentially expand the use of PaLM to more privacy-focused scenarios, but Google has not committed to a timeline for this development.
In the News: Android 14 combines generative AI and phone customisations
PaLM 2 in Bard
Google is making its Bard AI chatbot more widely available and adding new capabilities. In March, Google started testing Bard in the US and UK. Now, it is releasing it in over 180 countries and territories. Bard is a rival to ChatGPT.
Google has upgraded the Bard chatbot to the new PaLM 2 large language model to provide smarter answers to user questions. PaLM 2 and GPT-4 are models trained on reams of text data and can come up with human-like responses to questions and commands.
Bard will soon include images in responses, and in the next few months, Google will make it easier to prompt the chatbot with images through the Google Lens tool. Bard will be able to bring information from Google Maps, Google Docs, Google Sheets, and Gmail into conversations. Bard will also be able to show responses with help from third-party tools such as Adobe’s Firefly generative AI service, which can create images inspired by people’s text descriptions.
PaLM 2 in Workspace
Google Workspace will soon allow users to generate full documents and fill out spreadsheets with AI. Google’s Help Me Write feature that can write essays and sales pitches will be included in Google Docs for all people.
Another example the company gave was automatically writing job descriptions for online ads by prompting a senior-level job description for a textile designer. Google also demoed the software, based on PaLM 2, filling out an entire spreadsheet with three made-up dogs and notes for a fictional dog-walking company. You can take the structure and put in your own data and details.
Google Slides can generate AI images inside the app. The company said that the new features will be available later this year for subscribers to a service called Duet AI.
PaLM 2 in Search
Google aims to improve their most prized possession, Search, with generative AI — giving people real-time search results, that are more customised to their queries with links to dig deeper if they need to research more.
When people search for a question, a snapshot of quick information scraped from the internet along with links for more context will appear below the search box. People will also see suggested steps and the option of asking follow-up questions following the snapshot. These snapshots are going to populate with even richer results when searching for products and will include relevant, up-to-date reviews, ratings, prices and product images.
In addition to the generative AI snippet powered by Palm2, Google is also adding a ‘perspectives’ filter in search results, which will display short-form videos, images, written posts and discussions from social media platforms and online forums. Creator information will also be displayed, including their name, profile photo and popularity of their posts.
While AI seems to be piloting Search in the I/O 2023 announcement, Google maintains that they are “committed to continue sending valuable traffic to sites across the web”. To consistently improve the generative AI-powered Search, Google has also announced a Search Lab experiment called SGE (Search Generative Experience), that’ll take feedback from participants — available on Chrome for desktop, iOS and Android.
Also read: Google introduces AI into Maps and Photos