The Untrainable
The Untrainable AI progress will not render most companies as thin wrappers absorbed by frontier models, leaving value only in compute and base models like those from Anthropic and Nvidia Agents have dramatically improved measurable tasks per MIT data showing 180% more code written and 30% more shipped, yet real value persists in unmeasurable private correctness, complex workflows, organizational trust, and accountability that benchmarks cannot capture Valuable opportunities remain in applications that deeply integrate with private company data, translate AI into domain-specific operations…
The excerpt is quoted from a public Telegram channel, which owns its text.