SAN FRANCISCO — Last June, Inflection AI was on a roll. The artificial intelligence startup founded by veterans of Google's famous DeepMind AI lab is working with Microsoft and tech billionaires Bill Gates, Reid Hoffman and Eric Schmidt to build a chatbot business. It had just raised $1.3 billion from Mr.
But within a year, the tide had changed. Co-founders Mustafa Suleiman and Karen Simonian left Inflection AI to join Microsoft. The company also currently has the rights to use its technology. Inflection AI is currently focused on helping other companies improve their AI tools.
This isn't the only case of AI hype coming back to Earth. Even after 11 months of public testing, Google's AI search tools continue to make mistakes and remain closed to most people. A new scientific paper undermines some of the fancy claims about the technology's capabilities. The AI industry also faces increasing regulatory and legal challenges.
A year and a half into the AI boom, there is growing evidence that the hype machine is slowing down. The sweeping warnings that AI poses an existential threat to humanity or will take everyone's jobs have all but faded away, as do the stories about how to trick chatbots into helping you summarize insurance policies or answer customer service calls. It has been replaced by technical conversation. Some once-promising startups have already fallen from grace, with a series of flashy products announced by the biggest players in the AI race, OpenAI, Microsoft and Google, changing the way people work and communicate with each other. I haven't made any changes yet. While money continues to pour into AI, few companies are making money from the technology, and it remains prohibitively expensive to build and operate.
Technology executives, engineers and financial analysts say the road to adoption and business success remains long, winding and full of obstacles.
“If you think of a mature market as a mature tree, we're still just the trunk,” says founder of Gretel AI, a startup that helps other companies create datasets to train their own AI. said Ali Golshan. “We are in the early days of AI.”
The tech industry isn't slowing down on its AI efforts. Globally, venture capital funding for AI companies rose 25% to $25.87 billion in the first three months of 2024 compared to the second three months of 2023, according to research firm Pitchbook. It became. Microsoft, Meta, Apple, and Amazon are all investing billions of dollars in AI, hiring PhDs, and building new data centers. Just recently, Amazon poured his $2.5 billion into Anthropic AI, bringing his total investment in OpenAI competitors to his $4 billion.
Companies like Microsoft, OpenAI, and Google This means that even though companies are slowly starting to charge for AI tools, they are still costly. Billions of dollars are spent developing and running these tools.
Google Cloud CEO Thomas Kurian said earlier this month that companies like Deutsche Bank, Mayo Clinic and McDonald's are all using Google tools to build AI applications. Google CEO Sundar Pichai also said on the company's recent conference call that interest in AI contributed to the increase in cloud revenue. But on the same conference call, Chief Financial Officer Ruth Porat said the company's investments in data centers and computer chips to run AI mean Google's spending this year will be “significantly higher” than last year. said to mean.
“We help customers move from proof of concept to large-scale deployment of AI,” said Oliver Parker, an executive on the Google Cloud AI team. He cited Discover Financial Services as an example. The company leveraged Google's AI to build tools for call centers that reduced the time each call took.
Microsoft has also touted interest in its AI tools, saying 1.3 million people are currently using GitHub Copilot, its AI coding assistant. He also offers his AI assistant for $30 per month to millions of Microsoft Office users around the world. However, the company remains tight-lipped about whether any of the tools offer benefits compared to the cost of running them. Like Google, the company is putting a lot of energy into letting customers use its cloud services to run AI apps.
“We know that AI requires a paradigm shift,” said Jared Spataro, corporate vice president of AI at Work at Microsoft. “This is different from traditional technology deployments where IT departments flip a switch. Companies identify areas where AI can really have an impact, and that’s where they need to strategically deploy AI.” there is.”
In October, OpenAI announced its own version of the app store. This app store allows people to create their own custom version of the popular chatbot ChatGPT, upload it to a public marketplace, and receive a reward from OpenAI if many people use it. 3 million custom GPTs were created, but the company has not yet disclosed whether any money has been paid out.
“These tools aren't widespread yet, and they're not even close to being widespread,” said Radu Miklaus, an AI analyst at tech research firm Gartner. He expects that to start changing soon. He said: “We're hopeful that applications will start to pick up steam this year.”
Beyond Big Tech companies, many startups are finding ways to make money with “generative” AI technologies such as image generators and chatbots. These include trying to replace customer service representatives, writing ad copy, summarizing doctor's notes, and even trying to detect deepfake AI images created by other AI tools. It also includes. They want to sell subscriptions to their AI tools by replacing employees or helping them be more productive.
Many computer programmers say they already use chatbots to help write routine code. Utilization among the rest of the workforce is also gradually increasing, according to the Pew Research Center. According to the research organization, as of February 2024, approximately 23% of U.S. adults said they used ChatGPT at work, compared to 18% as of July 2023.
“We're still at the very beginning,” said John Yue, founder of Inference.AI, a startup that helps other technology companies find the computer chips they need to train their AI programs. says Mr. AI will permeate every industry, but it may take at least three to five years before people actually see the changes in their lives, he said. “We have to take a longer view.”
There are also major obstacles that can slow down the industry. Governments are also buying into the AI hype, and politicians at home and abroad are busy debating how to regulate the technology. In the US, smaller AI companies have expressed concerns that AI leaders like Google and OpenAI will lobby governments to make it harder for new entrants to compete.
In Europe, the final details of the EU's AI law are being worked out, promising to be more restrictive than what companies are used to seeing. One of the biggest concerns that AI data startup CEO Golshan has heard from business clients is that new AI laws will wipe out future investments. A series of lawsuits have also been filed against OpenAI and other AI companies for using people's work and data to train AI without payment or permission.
Although the technology continues to advance, there are still glaring issues with generative AI. Researchers have struggled to find ways to ensure that supposedly reliable models aren't fabricating false information. At Google's big cloud computing conference in early April, the company proposed a new solution to this problem. Customers who use the company's technology to train AI models will be able to have their bots fact-check just by searching on Google.
Some claims about AI's near-magical ability to perform human-level tasks have also been called into question. A new paper by researchers at the University of Massachusetts, Adobe, the Allen Institute for AI, and Princeton shows that AI models routinely make factual errors and omissions when asked to summarize long documents. Another recent paper suggests that claims that AI can outperform most humans at writing bar exams are overstated.
The huge improvement in AI technology demonstrated by ChatGPT, which started the boom, came from OpenAI feeding AI algorithms with trillions of sentences from the open internet. Subsequent AIs from Google, OpenAI, and Anthropic added even more data from the web to further improve capabilities. These improvements have led some prominent AI researchers to bring forward their predictions for when AI will surpass human-level intelligence. But AI companies are lacking data to train their models, raising questions about whether the steady increase in AI capabilities is reaching a plateau.
Training bigger and better AI models requires another key element. That's the electricity that powers the warehouse of computer chips that process all the data. The AI boom has already started a wave of new data center construction, but it's unclear whether the U.S. will be able to generate enough electricity to power them. Mike Hall, CEO of renewable energy management software company Anza and a 20-year veteran in the solar industry, believes that AI and the proliferation of new manufacturing facilities will drive the power needs over the next five years. He said it is pushing up volume forecasts. power industry.
“People are starting to talk about a crisis, but do we have enough power?” Hall said.
Ethan Mollick, an associate professor at the University of Pennsylvania who studies AI and business, said early experiments and tests have recognized the benefits AI can bring to their businesses, and executives are now exploring ways to further embed AI into their organizations. He said he was looking into it. Wide.
Some are hesitant to make large investments right now because they fear their AI tools will be rendered obsolete by new technology released in the future. OpenAI plans to unveil updated AI models in the coming months, which could offer an entirely new set of features, Mollick said.
“The underlying technology heartbeat hasn't really stopped,” he says. “It will take time for it to spread.”