Google announced updates to its Gemini AI models this week. The company released two new production-ready models, Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002. Google reported improvements in overall quality, especially in math, long context handling, and vision tasks.
https://twitter.com/GoogleDeepMind/status/1840710919044665823
Performance benchmarks show a 7 percent increase overall and a 20 percent improvement in math-related tasks. Google also cut prices for Gemini 1.5 Pro.
https://twitter.com/googlecloud/status/1839741802590441932
Input token costs were reduced by 64 percent and output token costs by 52 percent for prompts under 128,000 tokens.
Rate limits were increased as well. Gemini 1.5 Flash now supports 2,000 requests per minute. Gemini 1.5 Pro can handle 1,000 requests per minute.
Ask OCTO is our version of an advice column—except the relationships we're looking to improve are the ones in your tech stack.
In this edition, experts from Google Cloud's Office of the CTO discuss the data and AI virtuous cycle ↓ https://t.co/a66bXGlEnF
— Google Cloud (@googlecloud) September 27, 2024
These changes aim to make it easier and cheaper for developers to build applications with Gemini. Meta also made AI news this week. The company announced the release of Llama 3.2 on Wednesday.
It’s a major update to Meta’s open-weights AI model lineup. The new release includes vision-capable large language models (LLMs) in 11 billion and 90 billion parameter sizes. It also has lightweight text-only models of 1 billion and 3 billion parameters designed for edge and mobile devices.
Google’s Gemini AI models improve
Meta says the vision models are competitive with leading closed-source models in image recognition and visual understanding tasks. The smaller models outperform similar-sized competitors in various text-based tasks.
The new models support long context windows of up to 128,000 tokens. They are available for free download, with some license restrictions. In other AI news, Google DeepMind announced a big advancement in AI-driven electronic chip design with its AlphaChip technology on Thursday.
The project began in 2020 and has evolved into a reinforcement learning method for designing chip layouts. Google has reportedly used AlphaChip to create “superhuman chip layouts” in the last three generations of its Tensor Processing Units (TPUs). TPUs are chips designed to speed up AI operations.
Google claims AlphaChip can generate high-quality chip layouts in hours. This would take weeks or months of human effort. The company has released a version of AlphaChip on GitHub, sharing the model weights with the public.
AlphaChip’s impact has gone beyond Google. Chip-design companies are adopting and building on the technology for their own chips. This development has sparked new research in AI for chip design.
It could potentially optimize every stage of the process from computer architecture to manufacturing.
