Usama Fayyad, the main vice-prevail of the AI and the data strategy in Northeastern, says that companies should start small when IA introduce their operations.
Speak to Northeastern’s Roux Institute in Portland, in Maine. Photo of Matthew Modoono / Northeastern University
With all the media threshing around chatbots like Chatgpt, Gemini and Claude, many business leaders feel in a hurry to quickly revise their operations using a generative AI.
It would be a mistake, said Usama FayyadMain vice-prevail for AI and data strategy at the Northeastern University.
Instead, it is preferable for companies to slow down, reduce their concentration and start small.
“It is the philosophy opposed to what Chatgpt do and all the others in a generative AI,” said Fayyad last week, speaking during the inauguration AI in Action Summit of Business On the northeast campus in Portland, Maine.
“They say essentially:” We want a single model which is a Know-it-all model. It can solve any problem. He speaks all languages. He knows all areas of science. This is not the right direction, “he said. “In fact, I think it is the opposite direction to the place where we have to go.”
The most popular – and most promoted – used chatbots are now large language models, which means that they have been trained on large data strips. These models are not ideal for the business world, said Fayyad because they are swollen and with a high energy intensity.
It is much more practical for companies to turn to smaller and tailor -made models that gain more and more land in the AI research space.
“Small languages models provide an incredible solution,” added Fayyad. “They are effective, at high speed, private – you do not send your data anywhere – and you can customize it, which is the most important part.”
“Do not ask:” What is the biggest model that I can afford to build? You should ask, “What is the smallest LLM I can get away with?” “”
But even smaller and tailor -made models are as good as the data on which they are formed. And it was true since the researchers started working with automatic learning technologies in the 1940s, Fayyad said.
In fact, many of the basic technologies behind the generative tools of today’s AI are similar to those of the researchers used decades ago. The biggest difference is not algorithms – it is the massive increase in data available for training models.
“Between internet and digital transformation, it is a completely different world 70 years ago,” he said. “This is what made the difference.”
Consequently, the companies that succeed in the integration of AI into their operations are experts in data capture and action, said Fayyad. They understand that their data sets are their secret sauce.
“Many companies, I don’t care about your small size, you have the type of data that an Openai, Microsoft or Google would kill to have access,” he said.
For many workers, a generative AI can feel confusing, fast and difficult to apply effectively to work. But to remain competitive, companies and employees must invest in the literacy of AI and update, said Fayyad.
“” Will AI replace my work? ” is the eternal question these days, “he said. “The answer is no, but a human using AI will replace your work. … It is important to use this kind of thing.”
However, it is important to understand that these technologies can also be used harmful. Fayyad underlined the proliferation of Deep Feltfakes as an example.
“Today, we create a world where you cannot trust or believe everything you read, see and hear,” he said. “In this world, where you depend if digital, how can you work when everything you see is no more trustworthy?”
To combat these problems, Fayyad highlighted the importance of security regulations and managers of managers under development in universities like Northeastern.
“We think it is a huge area to think-how should we use this technology and where should we take it?”
