Google answers with PaLM 2 and a global Bard

At its I/O developer conference, Google announced PaLM 2, a new large language model family succeeding the original PaLM, and moved its Bard chatbot onto it the same day. Google said PaLM 2 was trained more heavily on multilingual text spanning over 100 languages, and had been strengthened for reasoning and coding, including proficiency in languages beyond mainstream ones such as Python and JavaScript. The model came in four sizes — Gecko, Otter, Bison and Unicorn — allowing deployment from mobile devices up to larger cloud workloads.

The more consequential change for users was to Bard’s availability. Google dropped the waitlist that had gated access since Bard’s limited March launch and opened the chatbot in more than 180 countries and territories, alongside support for Japanese and Korean with roughly 40 languages promised to follow. The company also announced a partnership bringing Adobe’s Firefly image-generation tool into Bard, along with visual search, export to Google Docs and Gmail, and a dark mode.

The announcement functioned as a direct response to Microsoft’s integration of GPT-4 into Bing three months earlier, and to the runaway adoption of ChatGPT more broadly: having launched Bard cautiously and to criticism over a factual error in its debut demo, Google used I/O to signal that its chatbot was now both more capable and open to a far larger audience. PaLM 2 underpinned Bard and other Google products through the rest of 2023, before Google replaced it with the multimodal Gemini family that December.