News from the AI & ML world
DeeperML
@cloud.google.com
//
Google is enhancing its Chrome security measures by integrating the on-device Gemini Nano large language model (LLM) to combat tech support scams. This new feature, launched with Chrome 137, adds an extra layer of protection by leveraging the LLM to generate signals that Safe Browsing can use to deliver more accurate verdicts on potentially dangerous sites. The on-device approach allows Chrome to detect and block attacks in real-time, even those from malicious sites that exist for less than 10 minutes. This method also considers how sites present themselves to individual users, enhancing the ability to assess the web for illegitimate purposes and potential threats.
AI Hypercomputer at Google Cloud is receiving several enhancements to accelerate AI inference workloads. These updates include the unveiling of Ironwood, Google's newest Tensor Processing Unit (TPU) designed specifically for inference, along with software improvements like simple and performant inference using vLLM on TPU and the latest GKE inference capabilities. With optimized software and powerful benchmarks, AI Hypercomputer aims to maximize performance and reduce inference costs, further enhancing JetStream and bringing vLLM support for TPU. JetStream, Google's open-source inference engine, has demonstrated significantly improved throughput performance for models like Llama 2 70B and Mixtral 8x7B.
Google is also investing in advanced nuclear power to fuel its AI and data center growth, emphasizing its commitment to sustainability and addressing the increasing energy demands of AI. Partnering with Elementl Power, Google plans to build three nuclear power plants, each generating at least 600 megawatts of clean electricity. These plants will utilize small modular reactors (SMRs), which are smaller, cheaper, and faster to build than traditional nuclear reactors, aligning with Google's goal to be pollution-free by 2030 and ensuring a constant, carbon emission-free energy source for its energy-intensive operations.
ImgSrc: storage.googlea
References :
AI Hypercomputer at Google Cloud is receiving several enhancements to accelerate AI inference workloads. These updates include the unveiling of Ironwood, Google's newest Tensor Processing Unit (TPU) designed specifically for inference, along with software improvements like simple and performant inference using vLLM on TPU and the latest GKE inference capabilities. With optimized software and powerful benchmarks, AI Hypercomputer aims to maximize performance and reduce inference costs, further enhancing JetStream and bringing vLLM support for TPU. JetStream, Google's open-source inference engine, has demonstrated significantly improved throughput performance for models like Llama 2 70B and Mixtral 8x7B.
Google is also investing in advanced nuclear power to fuel its AI and data center growth, emphasizing its commitment to sustainability and addressing the increasing energy demands of AI. Partnering with Elementl Power, Google plans to build three nuclear power plants, each generating at least 600 megawatts of clean electricity. These plants will utilize small modular reactors (SMRs), which are smaller, cheaper, and faster to build than traditional nuclear reactors, aligning with Google's goal to be pollution-free by 2030 and ensuring a constant, carbon emission-free energy source for its energy-intensive operations.
ImgSrc: storage.googlea
Share:
References :
- security.googleblog.com: Using AI to stop tech support scams in Chrome
- Compute: From LLMs to image generation: Accelerate inference workloads with AI Hypercomputer
- thetechbasic.com: Google invests in advanced nuclear power to fuel AI and data center growth
Classification:
- HashTags: #GeminiNano #AIHypercomputer #CleanEnergy
- Company: Google
- Target: Users
- Product: Gemini
- Feature: AI Hypercomputer
- Type: AI
- Severity: Medium