The dialog at Google’s annual developer convention, I/O 2024, on May 14 was dominated by AI. While Google shared updates on its AI fashions, notably Gemini, it additionally introduced new merchandise and options. These span from generative AI fashions with artistic output to integral {hardware} required to coach AI fashions. Here are a few of the new merchandise launched at the occasion:
GENERATIVE AI PRODUCTS
VEO
Google introduced the launch of Veo, a text-to-video AI mannequin able to producing high-definition video, developed by Google Deepmind. Veo is able to creating movies with 1080p decision based mostly on textual prompts. According to Google, Veo is able to “precisely capturing a immediate’s tone and rendering particulars in longer prompts.” VEO may understand cinematic phrases like “timelapse” or “aerial pictures of a panorama” and incorporate them into its output. It is offered for choose creators in non-public preview on VideoFX , Google’s new experimental instrument for video era. Google additionally plans to broaden Veo’s capabilities to YouTube Shorts and different merchandise.
Imagen 3
Google additionally launched Imagen 3, a text-to-image mannequin able to creating photorealistic pictures. Google added that it had “far fewer distracting visible artifacts than our prior fashions.” It can be able to rendering textual content, which Google acknowledged as “a problem for picture era fashions.” Imagen 3 is offered for choose creators in non-public preview in ImageFX, Google’s AI-generated instrument for text-to-image output. Based on buyer requests, Google has added picture modifying controls and higher-quality pictures to ImageFX permitting customers so as to add, take away or change particular components in pictures by merely brushing over them.
MUSICFX
Google shared that via its partnership with YouTube, it was collaborating with artists to discover generative AI in music. Google additionally shared that it was growing a set of music AI tools referred to as Music AI Sandbox. Further, it added a DJ Mode to its generative AI instrument MUSIC FX which might permit customers to combine beats by combining genres, devices, and so on.
WATERMARKING TECHNOLOGY
Google additionally plans to introduce watermarking capabilities for AI-generated textual content and video, via its watermarking instrument, SynthID. Launched final 12 months to watermark AI-generated pictures, the instrument was created to curtail the unfold of misinformation via AI. The SynthID watermark, permits customers to determine if the content material is AI-generated. Google mentioned that the watermarking know-how for artificial movies embeds a watermark instantly into the pixels of each video body, constructing on the approach used for watermarking pictures.
However, Google famous watermarking’s limitation saying that it was not a “silver bullet” however “an essential constructing block for growing extra dependable AI identification tools and will help hundreds of thousands of individuals make knowledgeable selections about how they work together with AI-generated content material.”
They additionally plan on making open-source SynthID for textual content watermarking and publishing a paper on textual content watermarking so builders can incorporate it into their fashions.
AI HARDWARE
Google introduced the sixth-generation model of its AI-specific {hardware}, Tensor Processing Units (TPU) referred to as Trillium. The new TPU goals to scale back latency and prices. Google mentioned that it had 4.7 occasions enhance in peak compute efficiency per chip, double the High Bandwidth Memory (HBM) capability and bandwidth, and Interchip Interconnect (ICI) bandwidth than earlier variations of the TPU. It can be 67% extra energy-efficient than TPU model 5.
Also Read:
https://www.medianama.com/2024/05/223-all-the-products-launched-at-google-i-o-2024/
