Interview, Fireside Chat, Conference Presentation
OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide Craze
- OpenAI is projected to reach 1 billion weekly active users by the end of 2025, a target missed four months into 2026, while aiming to go public later in the current year.
- Despite missing consumer targets, the company expects product improvements to succeed, with coding identified as a critical growth sector anticipated to perform exceptionally well.
- Current compute constraints are less severe for OpenAI than for Anthropic, allowing service delivery despite lower user numbers, though future power and compute limitations could disadvantage both frontier models while benefiting hyperscalers like Oracle, Amazon, Meta, Microsoft, and Google.
- Hyperscalers' capital expenditure is projected to exceed $700 billion in 2026 and potentially reach $1 trillion within the next year, driven by persistent demand for tokens and intelligence.
- The AI sector is expected to contribute 75% of GDP growth in the last quarter, supported by massive infrastructure build-out, and will become synonymous with the American economy to fund social programs and national defense.
- Forty percent of announced AI projects are expected to be cancelled due to negative public perception and weak "halo" effects.
- Within approximately six months, all frontier AI models are expected to possess "Mythos-level" cyber capabilities, with Chinese models like DeepSeek V4 reaching full frontier capability relative to American models in the same timeframe.
- Over the next five to six years, cyber warfare is predicted to shift from humans exploiting humans to machines exploiting machines, accompanied by a massive upgrade cycle in cyber capabilities and a rewriting of all operational software to address inherent insecurity in "handmade" code.
- AI coding practices are expected to evolve from "vibe coding" to professionally managed software requiring human supervision to prevent data destruction, moving from the peak of inflated expectations toward a slope of enlightenment.
- Pruning techniques are anticipated to reduce neural network model sizes by 90% while maintaining accuracy, potentially lowering inference costs by 10x, with Small Language Models and Verticalized Small Language Models expected to handle the "long tail" of user queries.
- Elon Musk's legal case against OpenAI may result in a technical victory with a settlement of $40 million or 10-20% of the company after dilution, rather than a structural reversal of the organization.
- Retatrutide is projected to be widely available by mid-2027, with FDA approval potentially occurring sooner or later in the current year, and priced as a premium "Mercedes" alternative to the lower-priced "Honda" Terzepatide for Medicare patients.
- The Supreme Court's ruling on the Monsanto/EpA preemption case carries a 50-50 probability of favoring either side, potentially redefining state versus federal regulatory authority and establishing precedents regarding Chevron doctrine overturns.
- Supreme Court public favorability currently fluctuates between 44% and 50%, with speculation that the court could expand from nine justices to 13, 21, or a significantly higher number in the future.
- The "dark fiber" scenario of the late 1990s is not expected to repeat in the AI infrastructure sector due to current demand pulling forward investment, and AI is predicted to create a productivity boom requiring human supervision rather than replacing humans entirely.
- Doomer arguments halting AI progress are expected to fail as stopping AI development is viewed as equivalent to halting the American economy, while the Supreme Court continues its tradition of strict decorum and legal interpretation.