Social media is full of examples of Google's new AI review product saying weird things, from putting glue on users' pizza to suggesting they eat rocks. The messy rollout means Google is scrambling to manually disable AI review for certain searches as different memes are posted, which is why users tend to ignore them immediately after they're posted on social networks. I see many disappearing.
It's an odd situation, because Google has been testing AI review for a year now — the feature was launched in beta in May 2023 as a search generative experiment — and CEO Sundar Pichai has said the company has More than a billion questions asked in that time.
But Pichai also says that Google has cut the cost of delivering AI answers by 80 percent over the same time period, “driven by hardware, engineering and technological breakthroughs.” It seems like this type of optimization would have happened very quickly before the tech was developed.
A founder of AI, who did not want to be named, said, “A company that was once known for being on the cutting edge and shipping high-quality products is now known for producing low-quality products. the edge.
Google continues to say that its AI review product largely provides “high-quality information” to users. “Many of the examples we've seen are unusual questions, and we've also seen examples that were doctored or that we didn't reproduce,” Google spokeswoman Meghan Farnsworth said in an email. could.” the edge. Farnsworth also confirmed that the company is taking “aggressive action” to remove AI reviews on certain questions “where appropriate under our content policies, and broader improvements to our systems using these examples.” have been started, some of which have already started.”
Gary Marks, an AI expert and professor emeritus of neural science at New York University, said. the edge That many AI companies are “selling the dream” that the tech will go from 80 percent accurate to 100 percent accurate. Achieving the first 80 percent is relatively straightforward because it involves estimating a large amount of human data, but the last 20 percent is extremely difficult, Marks said. In fact, Marks believes that the last 20 percent may be the hardest part.
“You actually need to do some reasoning to decide: Is this plausible? Is this source legitimate? You have to do things like a human fact-checker might do,” Marx said. , which may indeed require artificial general intelligence, and Marks and MetaKA AI chief Ian Likon both agree that the larger language models that exist in existing AI systems such as Google's Gemini and OpenAI's Those who power the GPT-4 will not be the ones who create the AGI.
Look, this is a tough place for Google. Bing made big on AI Before Google did with Satya Nadella's famous “We made them dance” quote, OpenAI is reportedly working on its own search engine, a fresh AI search startup already has. It is worth it. $1 billion, and a younger generation of users who just want the best experience are switching to TikTok. The company is clearly feeling the pressure to compete, and the pressure itself leads to messy AI releases. Marks explains that in 2022, Meta released an AI system called Galactica that had to be removed shortly after launch because, among other things, it made people eat glass. Listen to the story.
Google has big plans for AI Review — the feature that's out there today is just a small piece of what the company announced last week. Multi-level reasoning for complex queries, the ability to create an AI-organized results page, video search in Google Lens — there's a lot to be desired. But right now, the company's reputation rests on getting the fundamentals right, and that's not looking good.
“[These models] They are constitutionally incapable of scrutinizing the seriousness of their work, and that is why the industry has had to be cut in the back,'' Marks said.