Jay Jay

The cost of waiting Why service-time variation and deadline-aware measurement determine usable transaction capacity. vi.ltd/update-2026-10-13-the…

What a transaction timestamp can establish Clock uncertainty, causal order and the capacity cost of verifiable timing vi.ltd/update-2026-10-26-wha…

I released a paper on quantum computing in 2017. My assessment then was that the hype was bullshit. Nearly a decade later, I remain unconvinced that the promises have been substantiated. Larger budgets, revised roadmaps and more polished announcements do not establish that we are closer to the capabilities being sold. “Closer” requires a measurable engineering argument connecting demonstrated performance to those capabilities. Repeating the destination is not evidence of progress towards it. The question remains: what can the system reliably deliver, under what conditions, and at what total cost? Until those answers justify the claims, I see no reason to withdraw my criticism. Nearly a decade is ample time to improve a sales pitch. Substantiating it is another matter.

Mathematics does not require a committee’s permission to be true. “Mathematicians did not ask for this work” is an extraordinary opening: nobody needed to commission the result. Every valid solution adds to our shared knowledge, whether it arrives by an approved route or inconveniences someone’s preferred vision of the subject. Trustworthiness matters enormously, but it rests on explicit assumptions, inspectable proofs and independent validation. A proof does not become invalid because its production lacked your blessing, any more than a press release makes it correct. “Ethically researched” needs considerably more precision than you have supplied. If you mean not plagiarising other people’s work, say so. If you mean accurate attribution, honest claims of originality, transparent methods or proper verification, name those obligations and identify the failures. AI-assisted work is certainly not exempt from them. But neither “ethical” nor “in the public interest” should become a decorative phrase covering whatever an association happens to approve. Ethics deserves an argument; it should not be reduced to a velvet rope around the mathematics department. The substantive issue is corporate governance. A Silicon Valley company choosing to release purported research without adequate controls raises questions about who authorised publication, what verification occurred, how contributions were credited and who takes responsibility for errors. Those questions apply to OpenAI as a company, just as they apply to any other research organisation. They establish no general indictment of AI as a tool. Hold the people making the decisions accountable, require the evidence and scrutinise the proofs. There is ample work there without appointing yourselves the licensing authority for discovery.