OpenAI Cofounder Declares ‘More Chips’ Won’t Fix AI’s Basic Common Sense Problem

KEY POINTS

•OpenAI cofounder Ilya Sutskever appeared on the 'Dwarkesh Podcast' in November 2025 to discuss AI industry trends.
•He challenged the belief that scaling hardware and data volume alone can transform AI progress.
•Sutskever stated data is finite and companies have amassed huge compute resources, requiring new research approaches.
•Now leading Safe Superintelligence Inc., he emphasized developing models that generalize better, similar to human learning.

In a world where companies pour billions into GPUs like kids hoarding candy, OpenAI cofounder Ilya Sutskever calmly dropped the mic on scaling hype during the November 2025 'Dwarkesh Podcast.' Sutskever, now leading Safe Superintelligence Inc., said relying on 'more chips, more data' is like thinking 100x more greasy pizza makes breakfast smarter. He scoffed at the last half-decade's 'low-risk, easy investment recipe,' warning that data is finite and giant data centers aren’t magic. Instead, he insisted it’s 'back to the age of research' where you actually invent stuff, but this time, with supercomputers doing homework.

Share the Story

(1 of 3)

Mockingbird News

@MockingbirdNews · now

Turns out, pouring billions into GPUs doesn’t magically AI-fy your tech. OpenAI’s Sutskever wants us back knitting algorithms, not just buying chips like they’re baseball cards.

194/280 charactersReady to post

Mockingbird News

@MockingbirdNews · now

When your AI strategy is 'buy more chips,' but the chips start looking back confused, it's time for actual research.

133/280 charactersReady to post

Mockingbird News

@MockingbirdNews · now

Safe Superintelligence Inc.: Because throwing money at GPUs is the adult version of hoping your plants grow by staring at them.

144/280 charactersReady to post