SoftBank CEO Says AI Designing OpenAI's Next Model
Treat this as discourse-shaping rather than disclosure: SoftBank CEO Masayoshi Son told CNBC that an AI "model is designing" OpenAI's next model, calling it a sign of "super intelligence" and shortening his superintelligence timeline from ten years to roughly two. Son said the claim came from direct conversations with OpenAI CEO Sam Altman and company engineers; CNBC quotes him saying "once that happens, [the] model generates [the] next model ... and it's going to be exponentially smarter than all of us." An OpenAI spokesperson declined to comment on unreleased models but noted AI is already used in parts of its development process. No technical detail accompanies the claim, nothing on whether "designing" means architecture search, data curation, or something broader, so there is nothing for practitioners to validate or act on. The signal is that a top OpenAI investor is publicly framing recursive self-improvement as already underway, which will shape capital allocation and safety debates regardless of its accuracy.
An unverifiable claim from a maximally interested party still moves markets and policy debates, which is why this one matters: SoftBank is one of OpenAI's largest backers, and its CEO is now publicly asserting that recursive self-improvement, a model designing its successor, has begun.
What was said
Masayoshi Son told CNBC that an AI "model is designing" OpenAI's next model, calling it a sign of "super intelligence." He said the information came from conversations with OpenAI CEO Sam Altman and company engineers, and was quoted saying, "once that happens, [the] model generates [the] next model ... and it's going to be exponentially smarter than all of us." Son also said his 2024 forecast of artificial superintelligence within ten years was deliberately "conservative" and that he now expects it within roughly two years. An OpenAI spokesperson declined to comment on unreleased models but noted the company already uses AI in parts of its model-development process. Seeking Alpha and Quartz carried the remarks; none of the coverage adds technical detail.
Editorial context
Automating parts of model design is an active engineering area: neural architecture search, AutoML, automated hyperparameter tuning, and distillation all reduce human effort in routine steps, and labs routinely use models to generate training data, evaluate outputs, and assist research engineering. That is materially different from autonomous end-to-end design of a frontier model, which would require combining architecture search, training-scaling decisions, data curation, and safety-aligned evaluation in one loop; no published result demonstrates that stack. Son's remark is an executive characterization, not a technical disclosure, and nothing in the coverage lets practitioners reproduce or validate the claim.
What to watch
Indicators that would substantiate the claim: papers or disclosures from major labs describing automated architecture or objective search applied to frontier-scale models; reproducible benchmarks showing model-discovered designs outperforming human ones; published evaluation frameworks for emergent capability and alignment; or OpenAI clarifying the actual scope of automation in its pipeline. Until then, the story's effect is on investment narratives and the safety debate, alongside Anthropic's calls to slow the pace of development, rather than on engineering practice.
Key Points
- 1SoftBank's Masayoshi Son told CNBC that an AI model is now designing OpenAI's next model, which he frames as emerging "super intelligence."
- 2Son says OpenAI engineers told him this directly and revised his superintelligence timeline from ten years to roughly two.
- 3The claim lacks technical detail and OpenAI declined to confirm specifics, so it shapes discourse more than practitioner action today.
Scoring Rationale
An influential investor publicly claiming, on the basis of conversations with OpenAI's CEO and engineers, that AI is now designing OpenAI's next model and revising his superintelligence timeline to about two years, a remark that meaningfully shapes industry and safety discourse. But it is an unverified verbal assertion with no technical detail, and OpenAI declined to confirm specifics, so its practical immediacy for practitioners is limited.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
