9 hours ago
We would like to be DeepSeek in the West, says Essential AI cofounder
Dearth of collaborative efforts in AI domain may widen inequality around these capabilities even as pressures to monetise AI models which are 'difficult to ignore' may eventually lead companies to prioritise less on a long-term focus in research, said
Ashish Vaswani
, CEO and cofounder at
Essential AI
, a San Francisco-based AI startup.
Vaswani, who earlier worked as a research scientist at
Google
Brain, is known for his pioneering contributions to the field of deep learning. In 2017, he co-authored the seminal paper 'Attention is All You Need' that broke new ground by introducing the transformer architecture which forms the foundation of generative AI applications like ChatGPT and its successors.
Vaswani later cofounded Adept AI labs with another co-author Niki Parmar, before co-founding Essential AI with Parmar in 2023. The startup, backed by the likes of Google, Nvidia and AMD, builds full-stack AI solutions that enhance efficiency for automating labour-intensive and repetitive work. Its mission is to 'deepen the partnership between humans and computers.'
Over the last few years, billions of dollars have been spent on scaling AI models, with less to show in terms of return on investments. Speaking to ET during this visit to India, Vaswani said this could impact AI research.
'There are several instances of companies shuttering the longer-term R&D efforts in interest of pouring all the resources into the money-making aspects, especially at a time of distress,' he said.
According to him, the onus is on the leadership to ensure healthy long-term bets in addition to building a sustainable business.
Discover the stories of your interest
Blockchain
5 Stories
Cyber-safety
7 Stories
Fintech
9 Stories
E-comm
9 Stories
ML
8 Stories
Edtech
6 Stories
Further, in the pursuit of leadership in the AI arms race, companies are running the risk of ignoring important ideas. 'Any race forces you to take existing ideas that work, and pour resources into it to scale it up, while ignoring other riskier, and alternative but faster paths,' Vaswani noted.
Lack of transparency
A lack of openness, unlike earlier, is hindering the advancement of ideas, according to Vaswani.
'In 2017, there were many labs that could produce revolutionary ideas, and models are just an artifact of that spirit. It is never a single innovation but a series that comes together and then gives you these nonlinear improvements. The chance of that being widely shared is lower today,' he said. The end goal of creating artificial general intelligence (AGI) too is narrowing people's view, according to Vaswani.
AGI is the stated aim of ChatGPT-maker
OpenAI
, for instance.
'I don't think of AGI as much as a state of progression, like growth in capabilities or intelligence growth. I don't think there is an end to these progressions,' said Vaswani.
Need for open-source
At Essential AI, therefore, the team is building an open science or open-source frontier model that everybody can control, said Vaswani. He explained that the idea is to put their work on frontier models out in the open so that others can build on them. This would include opening up its weights, and even to a certain extent data that they are using for training, he added.
'I think this is a strategic imperative because the point when we are able to reach the same level in the domains we care about, which is frontier, companies will trust our model more because they can see the process,' he said. On generating returns on investments from using open-source models, Vaswani said, 'I would not say it is trivial to make money as a closed-source model today, either - they are spending a lot of money, right?'
He sees a growing market
'Having spoken with potential future collaborators and customers, they would love to use an open-source model in collaboration with us. I think there are business opportunities. I don't want to cast this moniker on us. But, in the short term, we would like to be the
DeepSeek
in the West,' he said, referring to the Chinese AI startup which rapidly gained prominence for developing high-performing and cost-efficient large language models.
