Sam Altman: Getting Fired (and Re-Hired) by OpenAI, Agents, AI Copyright issues

Altman said OpenAI may not even call its next major model “GPT-5,” because he expects AI systems to improve continuously rather than through clean numbered jumps. He pointed to GPT-4’s steady gains as the better preview of that future, and said this pattern may also be easier for society to absorb.

OpenAI is deliberately trying to build a business moat around the entire “intelligence layer,” not merely the smartest model weights. Altman said lasting value will come from product, stickiness, price and surrounding systems—an implicit answer to open-source models that can approach frontier performance.

Altman’s preferred AI assistant is not an alter ego but a separate, senior-employee-like agent that knows context, acts for you and anticipates needs. Crucially, he wants it to reason, push back, warn about consequences and sometimes resist instructions rather than behave like a sycophant.

OpenAI has deliberately held back from building a music model because Altman thinks artist rights become especially difficult when systems imitate identifiable styles. He expects the fight to shift from training data toward inference-time permissions and payment, with opt-in/opt-out and economic rules for prompts like “in the style of Taylor Swift.”

Altman floated “universal basic compute” as a possible successor to traditional UBI: people could receive a transferable share of future AI compute instead of cash. He imagined users spending it, reselling it or donating it to work such as cancer research—effectively giving everyone a slice of AI productivity.

Altman said he learned he was being fired while in a Las Vegas hotel for F1 weekend; within hours he was considering simply returning to AGI research. Board members called the next morning about bringing him back, and despite condemning their decisions, he said he never doubted their integrity or commitment to safe AGI.

Altman said he regrets not taking OpenAI equity because the unusual arrangement fuels suspicions about his motives and outside deals. He said proposed device or chip-fab ventures would give equity to OpenAI rather than himself, and said he genuinely did not know where the reported $7 trillion figure came from.

Friedberg described AlphaFold 3 as extending protein prediction to interactions involving small molecules, potentially screening drug side effects in software before trials. He also noted Alphabet kept the core IP inside Isomorphic Labs, while outside scientists received a non-commercial web interface rather than open-source commercial access.