Google’s long-awaited Gemini model shows that the technology is losing its magic.
After nearly seven months of rumors and delays, Google has finally released its most advanced generative-AI model to date: Gemini 1.0, a program the company is advertising as one of the most capable pieces of software ever. It can purportedly solve calculus problems, explain memes, write code, and—in a real example offered by the company—provide feedback on cooking photos to help you decide when your omelet is done. Google is even billing Gemini as “a first step toward a truly universal AI model,” one that is designed from the ground up to engage with images, video, text, audio, and computer code in a range of contexts. And, somehow, it all feels a bit underwhelming.
Perhaps that is because today’s announcement feels like any other Silicon Valley product launch. Gemini comes in three different versions—Nano, Pro, and Ultra—suited for tasks of increasing complexity, akin to an iPhone 15 Plus, Pro, and Pro Max. (Nano and Pro are available now, though Ultra won’t be out until early next year; for now, it’s a branding exercise.) At the highest end, Gemini can outperform OpenAI’s top model, GPT-4, on most metrics, but these are iterative advances—less like the invention of a smartphone and more like adding a few megapixels to its camera. Gemini’s launch provides the clearest sign yet that generative AI has reached its corporate phase.
Read: Money will kill ChatGPT’s magic
Gemini’s release comes exactly one month after Google’s top chatbot competitor, OpenAI, hosted its first developers’ conference—a day dedicated to showcasing new GPT-based products and kickstarting the associated revenue streams that looked an awful lot like Apple’s annual Worldwide Developers Conference or Google’s similar I/O event. OpenAI’s DevDay helped precipitate an epic clash at the start-up’s San Francisco headquarters, in which the CEO Sam Altman was ousted, at least in part for commercializing the company’s technology too rapidly. He was reinstated five days later with the backing of OpenAI’s most powerful investors. The biggest of those backers, Microsoft, now has a nonvoting seat on OpenAI’s nonprofit board, adding the world’s second biggest company to an entity that had initially been structured to beat back corporate greed.
Google, OpenAI, and Microsoft aren’t alone in the corporate generative-AI race. On an earnings call earlier this year, Meta’s CEO, Mark Zuckerberg, explicitly said that improving AI translates to more engaging products and, in turn, more advertising dollars, and that so-called foundation models such as Gemini, or Meta’s own Llama, are creating “entirely new classes of products.” Late last month, Amazon released its own AI-powered assistant, Amazon Q, aimed at helping businesses, and a recent Amazon earnings report touted a “slew of generative AI releases” that would bolster the company’s profits. The AI boom turned the chipmaker Nvidia into a trillion-dollar company. Even Apple, late to the party, is reportedly spending millions of dollars a day to develop its most advanced AI models and is starting to release software frameworks that would allow developers to incorporate generative AI into apps. At the moment, generative AI is more about competition than revolution.
Read: The future of AI is GOMA
As Big Tech vies for the same pot of gold, AI products are becoming hard to distinguish. In a note accompanying Gemini’s release, Google’s CEO, Sundar Pichai, said that generative AI will accelerate knowledge and productivity in ways “we haven’t seen before.” But we have heard claims that the latest smartphone, VR headset, or office software will do the same—again and again, for years. Google’s chatbot, Bard, which uses the Gemini Pro model as of today, has gone through three underlying generative-AI programs in the past year; OpenAI’s ChatGPT has also received a few software updates in the past year. Bard and ChatGPT learned to describe images; the rates at which they misled and hallucinated decreased; their prose got tighter and more direct. But the fundamental experience of trying to tease useful responses from a quasi-intelligent screen hasn’t changed, just as driving a car that gets better mileage isn’t appreciably new so much as a tad better.
These AI models have also started to provide gateways into whole ecosystems of other products, not dissimilar to the way that the iPhone has features that crossover with the MacBook. Bard can help you out in Gmail, Google Drive, and on YouTube, and GPT-powered AI assistants litter Microsoft’s own suite of cloud services. Nano is expressly designed to run on smartphones, and Google is incorporating it into its Pixel 8 Pro device. The entire Gemini suite is also optimized to run on custom-made computer chips that Google advertises as the best way to train AI. Microsoft has taken a similar approach with its own models and chips; every step of the AI pipeline involves a Big Tech product, whether for people using AI software or developers coding those programs on Big Tech’s data servers. The tech giants have much to gain from productizing generative AI and its spin-off software: It is one component of many that helps lock consumers into a company’s digital universe.
The profit motive, now more explicit than ever, may justify even more secrecy from these tech companies. Apple fought for years against laws that would help individuals open up and repair their gadgets, in part on the basis that valuable trade secrets would be revealed. Keeping a tight lid on AI—already an opaque technology—might be justified along the same lines.
Gemini supposedly received extensive evaluations for “factuality, child safety, harmful content, cybersecurity, biorisk, representation and inclusivity.” Yet the public knows very little about those checks, akin to how the harmful content-moderation and labor practices of earlier software and gadget giants were kept under wraps. (Google did not immediately respond to a request for comment about Gemini’s safety evaluations.) That inscrutability makes it hard to verify claims about AI’s capabilities, let alone to prevent the building of models that use, say, exploited labor or pirated material. It is a fair bet, then, that whether the generative-AI race prompts genuine societal transformation or simply provides a new profit model, its winners will be whichever tech companies best execute Silicon Valley’s decades-old playbook.
Generative AI’s iPhone Moment
38
1
07.12.2023
Google’s long-awaited Gemini model shows that the technology is losing its magic.
After nearly seven months of rumors and delays, Google has finally released its most advanced generative-AI model to date: Gemini 1.0, a program the company is advertising as one of the most capable pieces of software ever. It can purportedly solve calculus problems, explain memes, write code, and—in a real example offered by the company—provide feedback on cooking photos to help you decide when your omelet is done. Google is even billing Gemini as “a first step toward a truly universal AI model,” one that is designed from the ground up to engage with images, video, text, audio, and computer code in a range of contexts. And, somehow, it all feels a bit underwhelming.
Perhaps that is because today’s announcement feels like any other Silicon Valley product launch. Gemini comes in three different versions—Nano, Pro, and Ultra—suited for tasks of increasing complexity, akin to an iPhone 15 Plus, Pro, and Pro Max. (Nano and Pro are available now, though Ultra won’t be out until early next year; for now, it’s a branding exercise.) At the highest end, Gemini can outperform OpenAI’s top model, GPT-4, on most metrics, but these are iterative advances—less like the invention of a smartphone and more like adding a few megapixels to its camera. Gemini’s launch provides the clearest sign yet that generative AI has reached its corporate phase.
Read: Money will kill ChatGPT’s magic
Gemini’s release comes exactly one month after Google’s top chatbot competitor, OpenAI, hosted its first developers’ conference—a day........
© The Atlantic
visit website
login
who are we?
contact us
qosheapp
Toi Staff
Gideon Levy
Belen Fernandez
Rami G Khouri
Donald Low
Mort Laitner
Ali Fathollah-Nejad
Nikkei Editorial
Michael Kwet
