Google quietly launches internal AI model named 'Goose' to help employees write code faster, leaked documents show
Hugh Langley Feb 15, 2024 https://www.businessinsider.com/google-goose-ai-model-language-ai-coding-2024-2 * Google has an internal large language model named Goose, designed to make employees more productive. * Goose is trained on "25 years of engineering expertise at Google." * The company is leveraging AI as part of a bigger efficiency push. As Google injects artificial intelligence into every product it possibly can, it’s also turning to AI to make its business more efficient. The company has launched a large language model named Goose that's just for employees. It is designed to assist with building new products, according to internal documents reviewed by Business Insider. The documents describe Goose as a “descendant of Gemini,” its big new large language model. Goose is “trained on the sum total of 25 years of engineering expertise at Google,” one of the documents states. “It can answer questions around Google-specific technologies, write code using internal tech stacks and supports novel capabilities such as editing code based on natural language prompts,” an internal summary of Goose reads. Leveraging AI internally could be a key factor in Google’s new efficiency drive, which has resulted in leadership cutting thousands of jobs over the past 13 months and a number of team re-orgs. Google finance chief Ruth Porat hinted at initiatives like this during the company’s recent earnings call, telling analysts that Google was working to “slow expense growth” through a variety of tactics including “streamlining operations across Alphabet through the use of AI.” One internal document notes that Goose is part of a plan to “bring AI to every stage of the product development process.” A Google spokesperson didn't respond to a request for comment. Goose appears to be available for at least some employees to use right now, however, it’s not clear if it’s fully functional in terms of its abilities. One document notes that Goose “is also planned to be the first general-purpose LLM approved for internal coding use at Google.” “Goose has a 28k token context window, which makes it particularly valuable for development tasks,” it adds. Token context windows refer to the amount of text, numbers, and other information a language model can take into account when processing a query. That's especially important for coding, which can include hundreds of thousands of lines of code. The documents note that the creation of Goose was a collaborative effort between Google Brain, DeepMind, and Google's internal infrastructure teams. The company merged Brain and DeepMind last year to build Gemini, which was in turn named after the twin teams that created it. Duck, Duck, Goose Google isn't the only tech company using its own AI models and products to boost productivity. Microsoft recently rolled out its Copilot tool to internal teams, Business Insider reported. And Goose isn't the only example of Google using AI to be more efficient. The company uses AI to try to reduce the amount of energy required to cool its data centers, for example. It is also increasingly pushing advertisers to use AI automated products for deciding where their money should be spent across Google's products. For now at least, Google insists that AI is not taking away employees’ jobs. “We’re not restructuring because AI is taking away any jobs,” said Google’s chief business officer Philipp Schindler, after Business Insider’s reported layoffs in Google’s ad sales team as part of a re-org. Goose could become increasingly useful for employees. The tool was designed with a "focus on new engineering-related capabilities," one of the documents reads, suggesting it could go well beyond the existing development tools offered to staff. And if Googlers have specific development questions while using Goose, they're encouraged to turn to the company's internal chatbot, named Duckie. Are you a current or former Google employee? Got something to share?? You can reach reporter Hugh Langley via encrypted messaging app Signal (+1 628-228-1836) or email ([email protected]<mailto:[email protected]>) -- _______________________________________________ Link mailing list [email protected] https://mailman.anu.edu.au/mailman/listinfo/link
