Skip to content
IFAI

AI models

Google is set to unveil PaLM 2 at I/O

Published Reach 44

Google is reportedly planning to launch its latest large language model, PaLM 2, at its developer conference this week. Internal documents show the model supports over 100 languages with advanced coding and math skills, alongside new generative experiences rolling out to Bard, Search, and Workspace. Why it matters: Google is ramping up its defense against Microsoft and OpenAI by aggressively injecting generative AI across its flagship products.

Google is set to unveil PaLM 2 at I/O
Image: cnbc.com

Google is finally bringing its heavy artillery to Google I/O.

According to internal documents viewed by CNBC, artificial intelligence will be the absolute centerpiece of the developer conference. Sundar Pichai is set to pitch the company's latest frontier updates to a live crowd, led by a brand-new foundation model called PaLM 2.

Unveiling PaLM 2 and a beefier Bard

The star of the show is PaLM 2, which has been operating internally under the codename Unified Language Model. The new model supports over 100 languages and has undergone extensive testing across coding, math, analysis, and creative writing.

Google first announced the original PaLM in April 2022 and rolled out an enterprise API this past March. The company also recently revealed Med-PaLM 2, a specialized medical model that can answer medical exam questions at an expert doctor level with 85% accuracy.

Meanwhile, Google's conversational bot Bard is getting a significant upgrade. Documents indicate Bard will gain expanded coding, math, and logic chops, alongside new language support for Japanese and Korean.

Behind closed doors, Google has also been cooking up even more powerful Bard flavors. Separate documents show the company tested a multimodal version called Multi-Bard, which relies on a larger dataset to crack complex math and coding tasks, alongside experimental builds dubbed Big Bard and Giant Bard.

Generative search and Workspace upgrades

Google is also aggressively embedding generative tech directly into consumer tools. Search and Bard will both see new generative experiences aimed at rethinking how people query information.

Workspace apps are getting a major AI collaborator boost as well. One screenshot viewed by CNBC shows a Google Slides sidebar with a chat box that lets users type a text prompt to create custom images directly on a slide. Google is also working on template generation inside Sheets, alongside features for Meet and Docs.

On the hardware front, Google will also show off its long-rumored Pixel Fold, which features water resistance and what Google claims is the most durable hinge on a foldable device.

Why it matters

Google has been on the defensive ever since Microsoft backed OpenAI and started injecting generative smarts into Bing and Windows. This showcase looks like Google's attempt to prove it can ship frontier-level tech without breaking its core products.

For everyday users and developers, having these capabilities natively wired into Gmail, Docs, and Google Search is a massive deal. It removes the friction of jumping between third-party chatbots and your actual work documents. The multi-language expansion to Japanese and Korean also gives Google an immediate global wedge.

The real challenge, in our view, is margin and compute. Rolling out massive models like Multi-Bard across billions of search queries is insanely expensive, and generating direct answers threatens Google's classic search-ad cash cow. PaLM 2 proves Google still has top-tier research muscle, but turning that into profitable, reliable user experiences is where the real fight begins.

Sources (1)

  1. [1]
    cnbc.com(opens in a new tab)Originalcnbc.com

Similar events

Past happenings like this one, for context.

Best of AI, daily