Google debuts PaLM 2, offering improved multilingual, reasoning, and coding capabilities, available in four sizes and powering 25 Google products and features
When you look back at the biggest breakthroughs in AI over the last decade … Sean Michael Kerner / VentureBeat : Google opens up about PaLM 2, its new generative AI LLM Aman Gupta / Livemint : What is Google's PaLM 2 language model and where is it being used? Emma Roth / The Verge : Google launches an AI coding bot for Android developers Frederic Lardinois / TechCrunch : Google says PaLM 2 is built on JAX and TPU v4, will power its updated Bard chat tool, and is available to developers via its PaLM API, Firebase, and on Colab Business Post : Google unveils new artificial intelligence-based tools James Vincent / The Verge : Google announces PaLM 2 AI language model, already powering 25 Google services Adam Conway / XDA Developers : Google could leapfrog ChatGPT with PaLM 2 in Bard Ina Fried / Axios : Google debuts PaLM 2, its latest model for AI Kyle Wiggers / TechCrunch : Google's PaLM 2 paper is forthcoming on many of the LLM's major limitations, but doesn't reveal which data or hardware setup the company used to train the model LinkedIn: Khash Nejad : PaLM 2 is available now! — We're introducing PaLM 2, our next generation language model. PaLM 2 is a state-of-the-art language model with improved multilingual … Mastodon: Ken Yeung / @thekenyeung@mastodon.social : Sissie Hsiao (VP & GM, Assistant) on stage. — Bard can now collaborate on code generation, debugging and explaining code snippets. Learned more than 20 programming languages. — starting next week: code citations will be improved. … Tweets: Jack Lowe / @mrjacklowe : “Google is attempting to reclaim its crown as the leader when it comes to harvesting copyrighted material from the internet for their own profit...” https://www.theguardian.com/ ... Shashank Joshi / @shashj : “In its preliminary research, the company [Google] warned that systems built on PaLM 2 “continue to produce toxic language harms”, with some languages issuing “toxic” responses to queries about black people in almost a fifth of all tests” https://www.theguardian.com/ ... Sundar Pichai / @sundarpichai : 2/ We also introduced our next generation model, PaLM 2. It's powering over 25 new products + features, has improved multilingual, reasoning and coding capabilities, and can be fine-tuned for specific uses, like health and security. https://blog.google/... Ray Wong / @raywongy : RIP book editors #GoogleIO https://twitter.com/... Steven Sinofsky / @stevesi : Comparing different non-deterministic, ever-changing chatbots to each other is going to generate a lot of non-reproducible conclusions. Looking forward to the arXiv “preprints”. @google : Today, we're introducing our latest PaLM model, PaLM 2, which builds on our fundamental research and our latest infrastructure. It's highly capable at a wide range of tasks and easy to deploy. We're announcing more than 25 products and features powered by PaLM 2 today. #GoogleIO @dharmesh : Watching Google I/O. Google is launching PaLM 2. This may be what we have been waiting for from Google for a while. Now waiting to see *when* devs can actually use it to build with. @nirave : Almost took a nap during PaLM 2 and Bard talk. Can someone use AI to tell me why I should care about these AI improvements? #GoogleIO https://twitter.com/... Robert Scoble / @scobleizer : Fresh meat for AI-first humans. https://twitter.com/... Ina Fried / @inafried : New: Google debuts PaLM 2, its latest model for AI, which now powers Bard chatbot and two dozen other products https://www.axios.com/... Dieter Bohn / @backlon : Bard knows over 20 coding languages https://twitter.com/... @google : We recently moved Bard to PaLM 2, a much more capable model. #GoogleIO Matthew Lynley / @mattlynley : Waiting for more details on this but looks like there will be multiple sizes for PaLM 2. No param counts mentioned but that Gecko can run on a local device. https://twitter.com/... @gsterling : Med-PaLM 2 outperforming doctors on medical exams #GoogleIO https://twitter.com/... @google : Med-PaLM 2 can help answer questions and summarize insights from a variety of dense medical texts. We're working to add more capabilities to Med-PaLM 2, so that it can synthesize information from medical imaging like mammograms or help radiologists interpret results. #GoogleIO
Context & Ripple Effects
Google had already begun moving Bard onto PaLM-family models to improve math and logic responses while signaling coding features. PaLM 2 turns that incremental product update into a shared model layer across Google services and developer tools.
The rollout also establishes the base for Bard’s broader distribution: it was made available across 180 countries and territories the same day, and later gained additional language support and EU-requested privacy features.
First-order effects
- Google can deploy one upgraded model across more than 25 products while giving developers direct access through the PaLM API, Firebase, and Colab.
- Bard users and developers gain stronger multilingual, reasoning, and coding functions; the move follows Bard’s earlier PaLM-based math and logic upgrade.
Second-order effects
- Google’s product teams and third-party developers can build on a common model and tooling stack, making API and cloud-platform distribution more consequential for AI adoption.
- The model’s documented toxic-language risks, including in some non-English languages, make quality and safety evaluation a practical constraint as multilingual availability expands.
Third-order effects
- This points to competition shifting from a standalone chatbot race toward control of the integrated model, developer platform, and product distribution stack.
- If broad model reuse continues, providers with large application portfolios may gain an advantage in collecting deployment feedback and turning model improvements into default product features—while concentrating the impact of model-level failures.
The trend: PaLM 2 is one instance of AI industrialization: frontier models are being packaged as shared infrastructure for consumer products and developer platforms.