Artificial intelligence is a trend that has come to change the digital world as we know it. Prompts are the new verses and chatGPT the gospel…. Or is it? Google’s arrival has been strong, thanks to its language model of the future: PaLM. Now, in second edition.
AI chats are becoming the present of internet-related work. Digital marketing queries, code improvements (and even rewriting it), images that would pass for professional photography… hundreds of applications without which, right now, we would not know how to work in the same way. So it has been with Google Bard and ChatGPT, the current rulers. We already know a lot about OpenAI, but what about Google and the language model behind it? We are talking about PaLM 2.
PaLM 2, Pathaways Language Model
An artificial language model is what is behind intelligent chats, the structure or “brain” that makes them work, that can maintain a realistic and human-like conversation.
The PaLM model is born from LaMDA, an earlier Google model that was born in 2021 and was the first language behind Bard. It was only present in a test environment and for testing among a lucky few.
PaLM v1 was trained to have conversations with anyone. Its algorithm analyzes what you type to it, understands its position in the sentence, admits indirect interpretations and then generates responses.
The goal of PaLM 2 is to turn Google Workspace, Google’s assistance tools, into a permanent, personal assistant.
PaLM 2 appears in the public sphere on May 10, 2023 during Google I/O. If PaLM 1 was three times more powerful than chatGPT-3? the new version will be able to compete against GPT-4. It will be available in four sizes (Unicorn, Bison, Otter, Gecko), depending on their performance. The smallest, Gecko, can run offline on mobile devices. We don’t want to imagine the power of Unicorn.
PaLM 2 features
- Multi-language. Trained in more than 100 languages, including idioms, slang, poems, riddles… passing language proficiency tests with A+.
- Ability to reason. Having been trained with scientific data, mathematical and logical webs, PaLM2 has managed to advance in aspects such as common sense.
- Programming ability. It can write in languages such as Python or JavaScript, and also generate more specialized code such as Prolog, Fortran or Verilog.
- Compute-optimal scaling. This language has been built with a technique that scales the size of the model and the size of the training dataset, making, when we use it, the server spends less and inference is faster.
How to test PaLM 2
Google’s most advanced language model is not only available in Bard (for the moment, in English-speaking countries). It will also be integrated in many more sites, starting with its own API. Make code, edit existing code, troubleshoot problems, collect data…
You will also be able to continue with your techie tasks on Vertex AI, the machine learning platform that allows you to implement AA models and AI applications. A fully customized roadmap for you to develop your own intelligent application. What a way to go to the next level.
Finally, Google Workspace will also benefit from this model. Imagine PaLM 2 as a personal assistant that takes notes, generates meeting summaries, organizes emails the way you prefer… and does it in Gmail, in Docs, in Presentations, in Drive, in Meet…. a way to have all your documents connected and know more about you, the way you work and your preferences.
Many of these services still have a waiting list… but soon they will be the real competition to ChatGPT. And we don’t know who can win… what is certain is that this battle will change (and improve) the tech sector in every way.
Google’s artificial intelligence applications have also touched the healthcare field. Med-PaLM 2 has managed to improve X-rays and mammograms by synthesizing the information received by the machines. In a beta environment… but with possibilities in the near future.
Leave a Reply