- Updated: November 26, 2025
- 7 min read
Ilya Sutskever’s Learning‑First Superintelligence: A New Path to AI Advancement
× English हिंदीગુજરાતી Home My Feed Remove AdsGet Premium Content Go Pro @₹99 Credit Cards Instant LoansApply Now Credit Score Refer & Earn ₹500Refer Now Finance Tracker Learning Curve Fixed Deposits Business Loan Markets Markets FII & DII Activity Corporate Action Research Advice Earnings Technical Trends Stock Deals Intraday Large Deals Bonds Webinar Pre Market Stock Scanner Sector Analysis Unlisted Shares Stock Updates PRO Stock Lists Monsoon Tracker Seasonality Analysis Crypto Dashboard Web Stories Tax Calculator Forum Gold Rate Silver Rate IFSC Code Finder Storyboard18 IPO Economic Indicators Economic Calendar Global Markets US Markets ETFs Indian Indices News News Tech/Startups Auto Research Opinion Politics Personal Finance EPF Guide Real Estate Portfolio Watchlist Personal Finance Personal Finance Loans Fixed Deposit Comparison Fixed Deposit Interest Calculator Moneycontrol – SecureNow Health Insurance ratings Mutual Funds Mutual Funds Performance Tracker Top ranked funds My Portfolio Top performing Categories Forum New Fund Offers MF Simplified Commodities Mutual Funds Gold Rate Silver Rate Currencies Cryptocurrency Media Live TV & Shows Video Podcast Photos Invest Now Trade like Experts Invest in Unlisted Shares Specials Pharma Industry Conclave Unlocking opportunities in Metal and Mining #MFSummit2025 Moneycontrol mutual fund summit 2025 Advanced Technical Charts International Get App Be a Pro Home PRO Markets News Editor’s Picks Tech India Learning Curve Portfolio Watchlist Commodities MF Personal Finance Forum Videos HomeArtificial IntelligenceTo beat OpenAI, Ilya Sutskever is betting on ‘learning-first’ superintelligence that grows like a human To beat OpenAI, Ilya Sutskever is betting on ‘learning-first’ superintelligence that grows like a human Ads Rather than releasing a finished system, SSI would deploy AI agents with exceptional learning capabilities that accumulate expertise on the job, like human workers joining an organisation Arun Padmanabhan November 26, 2025 / 13:10 IST Story continues below Advertisement Remove Ad Ilya Sutskever has a controversial take: the way everyone thinks about artificial general intelligence (AGI) is wrong. The OpenAI co-founder and architect of GPT-3 isn’t trying to build an AI that can immediately perform every human job. Instead, his new company, Safe Superintelligence (SSI), is pursuing something different — a “superintelligent 15-year-old”. Story continues below Advertisement Remove Ad “I produce a superintelligent 15-year-old that’s very eager to go. They don’t know very much at all, a great student, very eager,” Sutskever said in a recent podcast.”You go and be a programmer, you go and be a doctor, go and learn.” The vision fundamentally reimagines how humanity reaches superintelligence. Rather than releasing a finished system, SSI would deploy AI agents with exceptional learning capabilities that accumulate expertise on the job, like human workers joining an organisation. “The deployment itself will involve some kind of a learning trial-and-error period. It’s a process, as opposed to you dropping the finished thing,” he said.Also read: AI hits a speed bump: Why the next big thing isn’t coming so fast Story continues below Advertisement Remove Ad This approach could prevent winner-take-all dynamics in AI. Sutskever predicts intense specialisation. “You’re going to have lots of different niches and you’re going to have lots of different companies who are occupying different niches.In this world, we might say one AI company is really quite a bit better at some area of really complicated economic activity and a different company is better at another area. And the third company is really good at litigation,” he said.Related stories Tech Mahindra to build 8 bn-parameter education LLM under IndiaAI Mission, targets March 2026 launchOpenAI projects 220 million paying ChatGPT users by 2030: Report After co-founding OpenAI in 2015 following his stint at Google Brain, Sutskever left the company in May 2024, worried that commercial priorities were overshadowing its safety roots. He went on to launch SSI the next month with Daniel Gross and Daniel Levy.When Gross joined Meta’s AI division in July 2025, Sutskever stepped in as the CEO. Why scaling won’t save us Story continues below Advertisement Remove Ad The shift comes as Sutskever declares the end of AI’s scaling era. “From 2020 to 2025, it was the age of scaling,” he said. “But now the scale is so big. Is the belief really that if you had 100x more, everything would be transformed? I don’t think that’s true. So it’s back to the age of research again, just with big computers.” Pre-training, the technique that powered ChatGPT, is exhausting the internet’s data. Simply making models bigger no longer guarantees breakthroughs. “There are more companies than ideas by quite a bit,” Sutskever noted. Scaling “sucked out all the air in the room” and caused everyone to pursue identical approaches. Now companies need genuine research insights, not just bigger budgets. The question is what those insights should target.Also read: The AI circular economy just got an upgrade: Inside the Microsoft, Nvidia, Anthropic deal The emotional gap Sutskever points to a fundamental mystery: why do humans generalise so much better than AI? He recounted a neuroscience case where a patient suffered brain damage that eliminated emotional processing. “He stopped feeling any emotion. He didn’t feel sad, he didn’t feel anger, he didn’t feel animated. He became somehow extremely bad at making any decisions at all.It would take him hours to decide on which socks to wear. He would make very bad financial decisions.” The lesson points to something missing in AI. “It should be some kind of a value function thing,” Sutskever said about emotions. “But I don’t think there is a great ML analogy because right now, value functions don’t play a very prominent role in the things people do.” He explained that current reinforcement learning requires models to take potentially thousands of actions before getting a score. “That means that if you are doing something that goes for a long time—if you’re training a task that takes a long time to solve—it will do no learning at all until you come up with the proposed solution.” “These models somehow just generalise dramatically worse than people. It’s super obvious. That seems like a very fundamental thing,” Sutskever added.Arun Padmanabhan first published: Nov 26, 2025 01:10 pm Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!Home Markets Loans SPIN2WIN News Portfolio Follow Us On: News more Business New Market New Stocks Income Tax Calculator Election Schedule 2024 India News Economy Mutual Funds Personal Finance IPO News Startups New USA Elections New Features more Gold Rate Today Silver Rate Today Trending News Startups New National News MC Videos MC You Tube Tools more SIP Calculator 2 SIP Calculator 3 SIP Calculator 1 New SIP Calculator 4 Specials more PowerYourTrade Master Your Money Game Changers Network 18 Sites more News18 Firstpost CNN News18 CNBC TV18 Forbes History TV18 Personal Finance more Home Loans up to 50 Lakhs Credit Cards Lifetime Free Finance Tracker Fixed Deposits Fixed Deposit Comparison Fixed Income Account Aggregator Quick Links more About Us New Contact Us Advisory Alert New Advertise with Us New Disclaimer Privacy Policy Cookie Policy Terms & Conditions Careers Financial Terms (Glossary) Analytical & Information Tool New Download MC Apps: Copyright © Network18 Media & Investments Limited. All rights reserved. Reproduction of news articles, photos, videos or any other content in whole or in part in any form or medium without express written permission of moneycontrol.com is prohibited.