“Frankly, what’s missing from these models today is that they don’t clearly acknowledge that they’re not sure about certain answers and that they need help,” Singh told AFP.
Since last year, Microsoft, Google and their competitors have been rapidly rolling out generative AI-based apps like ChatGPT, which generates content of all kinds based on simple, everyday language prompts and makes users believe it knows everything.
But despite all this progress, they sometimes fall into the abyss of “hallucination”, that is, inventing answers.
This weakness is an important problem to solve, says Vic Singh, who is responsible for Copilot applications at Microsoft, because customers of the American technology giant’s AI assistant will not accept information being fabricated, even if rarely.
“Very smart people are working on this problem… so that models can identify any questions they don’t know the answers to and ask for help,” he explains.
He stresses that a more modest model would be no less useful.
He gave an example of an AI-based customer service assistant. “I talked to a company that was spending $8 per query every time an agent answered it,” he says.
Although in half of the cases the machine has to call a person to find the solution, the company continues to “achieve huge savings.”
productivity
Vik Singh started at Microsoft in January, and in the summer took over leading the development teams for Copilot (AI-powered assistants) in sales, accounting, and services.
These less common commercial applications allow tech companies to justify their massive investments in generative AI because they are starting to generate revenue.
According to the Windows Group, Copilot conducts vendor research, saving time on customer contacts. Lumen Communications “saves about $50 million a year,” Singh says.
The startups behind the generative AI revolution promise highly advanced systems that could help “advance humanity,” says Sam Altman, CEO of OpenAI (funded largely by Microsoft).
But at the moment, new technology is mainly used to increase productivity and therefore profitability.
Vic Singh’s teams are working to integrate CoPilet directly into the group’s software and make it more independent, with less need for users to request it directly.
“For example, if you are a salesperson, two weeks after a conversation with a customer, an AI-based assistant could encourage you to follow up on the call, or better yet, send an automated email on your behalf, because the task was pre-approved,” the official explains.
employment
Before finding a solution to global warming, artificial intelligence is supposed to free humanity from boring and repetitive routine tasks.
“This is just the beginning,” Singh says. “We’re focusing first on the easiest applications” to implement.
The question remains: Will all these productivity gains translate into job losses?
Corporate executives, including K. Krithivasan, the chairman of Indian IT giant TCS, believe that generative AI will make call centers virtually obsolete.
But Vic Singen, like many Silicon Valley executives, is counting on technology to make people more creative or even create new jobs.
He cites his experience at Yahoo.com in 2008, when dozens of editors were tasked with selecting articles for the home page.
“We had the idea of using AI to improve this process, and some people wondered what would happen to the employees,” Singh says.
The automated system allowed content to be updated more quickly, increasing clicks on links, but it also created a need for new articles. “Ultimately, we had to hire more editors,” the executive concludes.
#Microsoft #official #programs #learn