There’s a lot at stake for Google to get this right as the tech giant aims to catch up with OpenAI
When Google $GOOGL dropped its new generative AI model Gemini in a prerecorded video demo last week, it highlighted the differentiating elements of the ChatGPT rival, such as its ability to converse out loud. But there seemed to be a gap between what was shown and the large language model’s true capabilities.
Sure enough, Google admitted the video was edited. In the YouTube description of the demo, it says “latency has been reduced and Gemini outputs have been shortened for brevity,” which makes it look like the AI system is responding faster than it is. The staged aspects of the demo got called out by sites like Bloomberg and TechCrunch.
Google may have taken note.
This week, the tech giant returned with two “live demos” demonstrating its new Gemini Pro tools, AI Studio and Vertex AI, which let developers build generative AI apps. AI Studio is a free, web-based tool that allows you to quickly create prompts to use when building apps. Vertex AI allows for customization of Gemini, such as fine-tuning it with a company’s data, and includes additional security features. Both tools, which were previewed for media outlets on Dec. 11 and 12, are available today.
Katie Camacho, a Google spokesperson, told Quartz that the demos of AI Studio and Vertex AI “were live during the virtual press conferences on Dec. 11 and Dec. 12.”
The change in approach underscores the need for Google to build confidence in its generative AI capabilities as OpenAI, Anthropic, and other rivals make waves in a fast-moving market.
At the start of the press briefing, Thomas Kurian, CEO of Google Cloud, said there would be “live demonstrations” before recapping the Gemini Pro announcements from last week.
Josh Woodward $WWD, vice president of Google Labs, presented different capabilities of AI Studio, such as using the platform to prompt writing real estate property descriptions after inserting photos from Google Drive. Once Woodward clicked the Run button, the tool took both the text and imagery and sent them over to Gemini, which generated a detailed description of the property, in the style of a real estate listing.
Later, Nenshad Bardoliwalla, director of product management of Vertex AI, gave a tour of what Vertex AI looked like and showed the different features of the platform. He highlighted, for instance, the studio’s language capabilities by translating a property description into Spanish. Another feature was the platform’s ability to utilize real-time web data, such as holiday rental prices. These functions were used to adjust the real estate-related prompt.
With OpenAI still in the lead, there’s a lot of pressure on Google to outperform against the AI startup. Google says that Gemini’s most advanced model beats OpenAI’s GPT-4, which came out in March, in seven of eight major industry benchmarks.
Now, Google will need to convince developers, who have already had access to OpenAI’s GPT-4 for some time, to use its new Gemini Pro tool. Developers have free access to Gemini Pro through AI Studio and Vertex AI, within limits, until general availability early next year, Google said. After, the company will charge per 1,000 characters or per image across both platforms.
“[O]ur goal is to attract developers with attractive price,” Kurian said during the press briefing.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
