Sam Altman Talks GPT-5, AGI, ChatGPT, OpenAI...

Sam Altman frames the GPT era as a once-in-a-lifetime scientific and technological shift—one that will be judged not by benchmark scores but by whether each new generation unlocks qualitatively new capabilities, including faster progress in discovering new science. He predicts that future GPT models (GPT-5, GPT-6, and GPT-7) will keep increasing “utility” and emerging properties, and he argues that the most meaningful milestone will be AI systems that can help humanity discover new science—potentially boosting the pace of scientific progress by orders of magnitude. In his view, this matters because technological shifts of this scale don’t arrive often, and waiting to get involved risks missing the moment when AI starts accelerating discovery rather than just automating tasks.

Altman also ties the speed of adoption to the product’s real-world usefulness. ChatGPT’s release is described as a rapid inflection point: early excitement among tech-forward users quickly turned into broader word-of-mouth, reaching a million users in about five days—far faster than Shopify’s own path to the same milestone. He credits viral growth to people actively telling others and trying it themselves, and he contrasts that with the slower, more typical adoption curve for major products.

A recurring theme is how public expectations have repeatedly missed the mark. Past predictions about AI’s trajectory—such as the idea that it would first replace physical labor, then gradually handle easier cognitive tasks, and only later (if ever) creative work—ended up reversing course. Altman uses that history to argue that the field’s “goal posts” keep moving: once models outperform expectations in one domain, skeptics adjust what counts as impressive. That pattern underpins his belief that AI could contribute to original science before advanced robotics arrive, especially because models can use code to verify claims.

The conversation also emphasizes how training data and modalities change what models can do. Altman highlights the importance of training on code, saying that learning to program improves reasoning—an effect he compares to how step-by-step work benefits people who use structured reasoning. He’s also excited about adding video to model inputs. While current systems can handle text (and, in the case of GPT-4, images), video would expand what models can learn and represent, potentially making the path to AGI faster than relying on language alone.

On risks, Altman acknowledges the seriousness of downsides but pushes for a more balanced focus on benefits. He argues that AI is delivering a rare, fundamental “toolbox reshaping” capability—bigger than the last major consumer technology shift, like the smartphone—because it increases what people can accomplish. He points to real adoption stories: doctors using AI to structure diagnoses and creatives integrating it into workflows.

Finally, Altman links optimism and “abundance” to how society integrates AI. He reflects on OpenAI’s earlier failures and dead ends before the language-model breakthrough, and he suggests that the interface itself—chat—makes powerful tools usable by children, older adults, and people who feel uncomfortable with technology. The discussion ends by stressing energy as a foundational constraint on progress, with the claim that many global problems ultimately reduce to energy—an argument that connects technological optimism to practical infrastructure choices.