news.volyx.in

Gemini 2.0: our new AI model for the agentic era (blog.google)

1015 points by meetpateltech · 602 days ago · 490 comments on HN

Article summary

Google has introduced Gemini 2.0, a new AI model designed for the 'agentic era' with native image and audio output and tool use. Gemini 2.0 Flash is available to developers and trusted testers, with wider availability planned for early next year. The model is built on custom hardware like Trillium, Google's sixth-generation TPUs. Gemini 2.0 is expected to enable new AI agents that bring Google closer to its vision of a universal assistant.

Main themes

  • AI models
  • Agentic era
  • Native image and audio output
  • TPU hardware
  • Cloud computing
  • On-device AI

What commenters say

  • The Gemini 2.0 model's native audio output is not yet available for general use, but its image generation capabilities show promise for tasks like inpainting and style transfer.
  • Some commenters argue that benchmarks for agentic tasks are needed, beyond just code reasoning, to truly evaluate AI models.
  • There is disagreement over whether on-device AI or cloud computing is more important, with some arguing that training compute is the key to success, while others believe inference costs will become a major factor.
  • The cost of running state-of-the-art models like chatgpt-o1 is prohibitively expensive for commodity hardware, requiring specialized servers and GPUs.
  • Some argue that Google does not need to win the on-device market, as inference can be handled by other companies, while others believe that Apple's upcoming on-device AI capabilities will force Google to respond on Android.
  • The development of inference chips is considered easier than training chips, and companies may be willing to sell their chips to Google or run its models on their platforms.
  • There is a debate over the economic viability of on-device AI versus cloud computing, with some arguing that the cloud is better suited for handling infrequent but computationally intensive tasks.
  • The comparison between Google's Gemini 2.0 model and other AI models, such as o1-pro, is seen as unfair due to differences in model size and capabilities.