Picture this – a super cool, smart tool that can understand and generate human-like text! That’s what Google just revealed at their I/O developer conference – a powerful tool called PaLM 2. This thing is like a big brother to Google’s chat tool, Bard, and is like OpenAI’s ChatGPT.
PaLM 2 is not just super cool, but also super handy. Developers can now use it through Google’s PaLM API, Firebase, and on Colab.
Now, you might wonder, ‘How did Google make this tool so smart?’ Well, Google didn’t share a lot of techy details about how they trained PaLM 2. They did mention it’s built on Google’s latest JAX and TPU v4 infrastructure, which are just some complex Google tech stuff.
What’s interesting is that Google believes that the size of the model (which is a whopping 540 billion parameters – imagine that!) isn’t really the most important thing. They think it’s more about what the tool can actually do, and boy, can it do a lot!
PaLM 2 is a real smarty pants. It’s great at reasoning things out like a human, solving math problems, and it even understands logic. How? Because Google trained it on a ton of math and science stuff, even including mathematical expressions.
It’s not just about numbers, though. PaLM 2 is also a coding whiz. It was trained on 20 programming languages – everything from popular ones like JavaScript and Python to old-school ones like Prolog, Verilog, and Fortran. It forms the backbone of Google’s specialized model for coding and debugging, Codey, which they’re also launching now.
Google also said that PaLM 2 knows more than 100 languages. So, it’s pretty good at understanding and generating text in multiple languages.
Google has designed PaLM as a family of models, each one focusing on different things. For example, there’s Med-PaLM 2 that’s all about medical knowledge and Sec-PaLM for security stuff. There’s even a smaller version that can work on smartphones!
Google’s been pretty careful about launching these AI features, saying they want to make these tools safely and responsibly. And, that’s the same with PaLM. We’ll have to see how well it does and how it handles tricky situations.
This article is based on a report by TechCrunch.
