Aidan Gomez is impressed with Deepseek, but I don’t think AI is Enterprise-Ready. In an interview with BusinessInsider, Cohere CEO said the company is looking for custom models. AI could be cheaper.
Cohere CEO Aidan Gomez felt that Deepseek’s case was proven by his belief that strong AI doesn’t have to be that expensive We have released a model that will blow a trillion dollar hole in the US stock market.
“I think we’ve tested the Cohere strategy we’ve been pursuing for a while,” the 28-year-old computer scientist said in an interview with Business Insider. “We don’t have to spend billions of dollars a year to produce competitive, first-class technology.”
But, former Google researcher Gomez considers Deepseek’s R1 a “really impressive release,” but he isn’t sure it should be a serious option for businesses.
He says that organizations are looking for customized AI models rather than off-shelves, and notes that they provide AI tools with access to sensitive data.
“We don’t see any companies that rely on R1 to sell to run their systems,” Gomez says. “We don’t see it as a competitor on our part.”
Why Deepseek is not enterprise-ready
Gomez has clear business reasons to make this claim as the leader of a $5.5 billion company building AI. However, as one of the eight Google Brain researchers who co-authored the original “Attention You Need” paper in 2017, the paper that sparked the generation AI boom, has made its position weight. I’ll carry it.
Also, some companies that some US companies are trying to adopt Deepseek are trying to rush into many of the troubles they’ve reported previously.
For Gomez, Deepseek is not a quick win for businesses, no matter how impressive its technology is.
“What we see from a company is that they don’t just want to buy a model,” he said. “We need to build something with that model. We need to deploy a lot of technical resources to see the value. It will take time.”
To unlock the “new layer of value,” he believes that companies need to carefully consider how they customize Core AI technology with their own data.
In November, in an industry-wide discussion about whether AI performance has improved, leaders cited private and synthetic data as key resources organizations must leverage to stay competitive. did.
That’s a point reflected in Gomez. And there are concerns that DeepSeek, a “backend by Chinese servers,” will ring out, US lawmakers are trying to ban startup software from government devices — Cohere CEO says if the model is increasingly sensitive , said businesses should put privacy first. data. “
“This unlocks corporate use because they are hesitant to build systems that are exposed to sensitive data,” he said. “Our competitors handle it in a less secure way.”
Deepseek has not returned requests for business insider comments about its data privacy policy.
Related Stories
All about AI agents
Gomez believes Deepseek’s R1 is impressive, but believes that the true value comes from transforming the base model into a tool that proves it will become another hot area for the industry this year. Agent AI.
Software programs that allow you to autonomously execute tasks are rising on the business leader agenda this year. Agent AI has been a hot topic in Davos, but Nvidia CEO Jensen Huang says that it will be the year 2025 will take off on the Consumer Electronics show.
Toronto-based Cohere, one of a handful of companies competing with AI rivals such as Openai, Google and Anthropic, focuses on leading AI agents into the company.
Last month, Cohere introduced North’s early access program, a proprietary agent AI designed to meet specific workloads.
Gomez sees it as another way to gain an edge over companies that want to deploy a base model like Deepseek’s R1.
Gomez refused to say how much it would cost to adapt the platform North For the needs of a specific company. He said that after “prepaid investments,” agents can “act autonomously” and allow companies to “start to enjoy value.”
Nvidia CEO Jensen Huang said the AI agent will take off in 2025. Getty Images
Deepseek The Disruptor
Despite his concerns about Deepseek, Gomez sees the startup as a positive disruptive force for the AI industry spun from Chinese hedge funds.
“The fact that they published their training efficiency figures allows people to see that they don’t have to be so capital-intensive to publish great models,” he said.
AI leaders continue to scrutinise Deepseek’s claim that they produced AI at a cost comparable to the performance of Silicon Valley’s best models. In the meantime, investors have questioned whether large AI infrastructure spending is still justified.
Addressing the implications of Sam Altman’s $500 billion Stargate project, Gomez said it was a mistake to “expend more and more” on the infrastructure to train AI models rather than “inference.” Inference refers to AI models that predict or determine new data, and training is the process of building the capabilities of a model.
“I think Deepseek is a big point of evidence,” he said.
Another meaning of the appearance of Deepseek is apparently nowhere or anywhere. This is a testimony of an open source approach.
There’s debate as to whether DeepSeek’s AI is really open source, but I secured a MIT license and made it an open weight for the model, but I haven’t disclosed the data I used to train it . -Sauce and closed sauce player.
Still, companies need more than an impressive Chinese model to build powerful AI into operations. As Gomez said, “Downloading the model isn’t enough.”
