Friday, July 12, 2024
- Advertisement -
More

    Latest Posts

    Indian startups should focus on building on top of foundational models: MeitY’s additional secretary Abhishek Singh says

    Indian startups should focus on building on top of foundational models as opposed to creating foundational models of their own, Abhishek Singh, the additional secretary of the Ministry of Electronics and Information Technology (MeitY) said during the Delhi NCR AI Alliance on May 17. “Foundational models are something that requires a lot of effort, a lot of money,” Singh said, adding that not every startup will be able to do that. “Plus the talent, that is, required, the skillset, that is, required to train a foundational model from scratch is very limited and it is in limited supply across the world,” he added. 

    The larger opportunity for startups will be in building on top of the foundational model that is developed in India. However, till it is developed, there are a lot of foundational models that are already available, which can be used for inference or fine-tuning and building applications that can be deployed, Singh explained. 

    Who will get access to GPUs under India’s AI Mission?

    Discussing the artificial intelligence (AI) mission that was approved in March this year, Singh said that all those startups and researchers who are building solutions using Indian datasets for the problem statement floated by the government will all get an opportunity to use the compute. “Prioritization will be of course the problem statements that we have identified, the health sector, education, agriculture, social impact but there might be other solutions that startups can think of which we may not also be able to,” he said. He added there would be sufficient compute capacity for all the country’s needs.

    Singh said that the request for proposal (RFP) for Graphics Processing Units (GPUs) will be made available as soon as the model code of conduct (MCC) is over. This RFP will be for the setting up of a supercomputer infrastructure that will have all the chips required by the industry. “I will not say it will only be NVIDIA, it will only be H100s, it can be AMD 300, it can be Gaudi 3, it can be a combination of the three,” he said, explaining that the compute provided will depend on the requirements of the users. He mentioned that the government is conducting a user survey and trying to figure out the needs of the users. 

    Even after the RFP is floated it will take some for such infrastructure to come. “It will take anywhere between 12-18 months to become actually operational,” he said. Till this infrastructure is in place, other players are setting up AI compute infrastructure and there will be a mechanism put in place that would allow those who need said infrastructure to get it at a subsidized rate. 

    Applying AI to Digital Public Infrastructure (DPI):

    Singh said that the kind of data being generated by the DPI applications offers great potential for building better solutions by adding an AI layer on top of it. “For example, for the agriculture ministry, whether they implement the PM, Kisan, or DBT [direct benefit transfer] for transferring money, or whether it’s for the crop research Kisan call center and all, they are generating a lot of data. Using that data, one can build applications on top of that which will ensure farmers can get advisories on voice in their mother tongue,” he explained.

    Most people are not able to use DPI applications in India because they are not comfortable navigating web pages or downloading apps. However, if the same people were given voice-enabled applications through natural language processing (NLP), it would be transformational, Singh said.

    How can the government address deepfakes?

    Singh explained that there is an international consensus to come out with technologies that can detect deepfakes. “Can we have provisions of labelling, watermarking, deepfakes, so the people should be able to know as also there are provisions where something, if an image or video is generated by AI, then to mention that this is AI generated?” Singh questioned. He added that of course, ensuring compliance for all of this would be challenging and ultimately the law will have to play a role in this. 

    Singh mentioned that there have been laws around misrepresentation since the time of Photoshop, so the same ways can be applied to AI as well with better training of police personnel and judicial officers for better enforcement. “If we just make regulations and don’t have the kind capacity or the means to enforce them, it would not mean much,” he mentioned. 

    Will Bhashini be impacted by OpenAI’s translation feature?

    Singh said that the introduction of Indian languages within OpenAI’s translation feature is a good thing. “This is going to happen. And as we move forward and as the compute power goes up, as the data sets go up, at the power of the models, models become much better, this will happen. So, Bashini will also need to kind of up their game well with other tools which are there,” he explained.

    Also read:


    STAY ON TOP OF TECH NEWS: Our daily newsletter with the top story of the day from MediaNama, delivered to your inbox before 9 AM. Click here to sign up today!


    The post Indian startups should focus on building on top of foundational models: MeitY’s additional secretary Abhishek Singh says appeared first on MEDIANAMA.

    Latest Posts

    - Advertisement -

    Don't Miss

    Stay in touch

    To be updated with all the latest news, offers and special announcements.