Interview, Fireside Chat, Podcast
Sam Altman: Getting Fired (and Re-Hired) by OpenAI, Agents, AI Copyright issues
- GPT-5 is anticipated to launch in summer, though its release format may differ from previous iterations and the name is not guaranteed; development is expected to shift from distinct numbered versions to a system of continuous improvement and retraining.
- OpenAI aims to reduce costs and latency dramatically in the near future to make advanced technology accessible to free users, with a goal of intelligence becoming cheap enough to meter and fast enough to feel instantaneous, though no specific timeline is provided.
- Strategic plans include opening source more models and creating a lightweight model capable of running on a phone, potentially around the 7 billion parameter range, while the company intends to maintain a lead through enduring product value rather than just model weights.
- Future trajectories suggest intelligence will be recognized as an emergent property of matter with multiple design approaches, huge algorithmic gains, and potential substrate breakthroughs to resolve compute, energy, and supply chain bottlenecks.
- Interaction paradigms are expected to evolve toward multi-modality and voice interfaces as the primary computing form factor, with generalizable reasoning systems capable of solving specific problems like protein modeling without full retraining.
- Ethical and economic expectations include addressing inference-time fairness, artist compensation for likeness generation, and the potential emergence of international regulatory agencies for safety testing similar to nuclear oversight, while current government rules like California's proposed framework are viewed as likely to become obsolete quickly.
- Risks include the expectation that frontier AI systems capable of significant global harm will emerge in the not-so-distant future, distinct from the current lack of material threat regarding recursive self-improvement or autonomous bioweapon design.
- Economic models may shift toward "universal basic compute" as training data value declines, alongside predictions that Apple faces a lack of innovation will due to missing creative will, and that Google's AlphaFold 3 will monetize via Isomorphic Labs to potentially unlock anti-aging and disease cures.
- Industry consensus anticipates a mixture of experts approach for reasoning, the dominance of scaffolding over raw model differences for differentiation, and a sold-out All In Summit featuring high-profile guests, while Apple's recent advertising regarding creativity is expected to be poorly received.