OpenAISam Altman on AGI, GPT-5, and what’s next — the OpenAI Podcast Ep. 1
At a glance
WHAT IT’S REALLY ABOUT
Altman maps AI’s next era: GPT-5, privacy, compute, devices
- Altman frames “AGI” as a moving target: systems already exceed many older definitions, and more people will claim “we have AGI” each year as capabilities climb.
- He defines “superintelligence” more concretely as AI that can autonomously discover new science or massively amplify human scientific discovery—seeing scientific progress as the main lever for improving lives.
- The conversation highlights practical capability leaps from agentic tools like Operator (with o3) and Deep Research, alongside challenges like model naming/versioning, user trust, and risks of overly agreeable personalities.
- Altman emphasizes compute and energy as core constraints, describing Project Stargate as a large-scale effort to finance and build unprecedented AI infrastructure, while also teasing longer-term bets like new AI-centric hardware with Jony Ive.
IDEAS WORTH REMEMBERING
5 ideas“AGI” is increasingly a perception milestone, not a fixed threshold.
Altman argues prior AGI definitions based on cognitive capability have already been surpassed, and that public consensus will shift annually as systems improve—even as the goalposts move.
Altman’s practical bar for “superintelligence” is scientific discovery.
He says a system that can autonomously generate new science—or dramatically accelerate human discovery—would feel “definitionally” like superintelligence and could unlock broad quality-of-life gains.
Agentic products create visceral ‘AGI-like’ moments for users.
Altman notes many people cite Operator using a computer (especially with o3) as their personal “AGI moment,” while Mayne points to Deep Research’s iterative web-following workflow as a step-change from summarization.
Reasoning models trade speed for solution quality—and users accept waiting.
Altman is surprised users will wait for better answers on hard problems, signaling a shift from “instant response” UX toward “deliberate computation” when value is clear.
Model releases are becoming continuous services, complicating naming and trust.
With ongoing post-training and frequent updates (as with GPT-4o), OpenAI is debating whether to keep one name (GPT-5) or introduce explicit versioning (5.1, 5.2) so users can choose preferred snapshots.
WORDS WORTH SAVING
5 quotesMore and more people will think we've gotten to an AGI system every year.
— Sam Altman
If we had a system that was capable of either doing autonomous discovery of new science or greatly increasing the capability of people using the tool to discover new science, that would feel like kind of almost definitionally superintelligence to me.
— Sam Altman
Memory is probably my favourite recent ChatGPT feature.
— Sam Altman
You cannot have a company like The New York Times ask an AI provider to compromise user privacy.
— Sam Altman
If people knew what we could do with more compute, they would want way, way more.
— Sam Altman
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome