At Google I/O on Wednesday, the tech giant announced a new large language model: PaLM 2(opens in a new tab).
PaLM 2 is the next iteration of PaLM (which stands for Pathways Language Model), which was launched in April 2022(opens in a new tab). It understands over 100 languages, has improved math, logic, and reasoning capabilities, and is pre-trained on Python, JavaScript, and other coding languages.
Google CEO Sundar Pichai announced four different sizes of PaLM 2: "Gecko," "Otter," "Bison," and "Unicorn." Gecko is agile enough to work on mobile devices even when offline. As of today, PaLM 2 powers 25 Google products including Bard, Workspace, and specialized versions for cybersecurity (Sec-PaLM 2) and medical knowledge (Med-PaLM 2). Developers can soon get their hands on the PaLM 2 API by signing up for the waitlist.
When the original PaLM was launched, it was on the cutting edge, and hailed for its ability to parse jokes(opens in a new tab). To say it was soon overshadowed by GPT-3 and GPT-4, the models powering OpenAI's ChatGPT, would be an understatement. It was clobbered. With PaLM 2, Google is introducing a new contender to challenge GPT-4.
Developers can soon get their hands on the PaLM 2 API by signing up for the waitlist(opens in a new tab).