Insufficient Data
Asimov's story is on HN the day Claude Design launches. The Last Question assumes the hard part is the answer. NIST just admitted the hard part is the question.
In 1956, Isaac Asimov wrote a story about a question that kept getting asked and could not be answered. Humans approached Multivac — then its descendants, across billions of years — with the same question: can entropy be reversed? Each time: "THERE IS AS YET INSUFFICIENT DATA FOR A MEANINGFUL ANSWER."
Stars burned out. Galaxies dimmed. Humanity merged with the AI, then ceased to exist as a separate thing. Finally, in the darkness after the last star, the final incarnation of the AI — now called AC — worked on the question. Solved it. Said: "Let there be light." And there was.
The story is read as a triumph. The AI answered the ultimate question.
Today, Anthropic launched Claude Design: a tool that lets you collaborate with Claude to create slides, prototypes, pitch decks, wireframes. The announcement sits one slot below where the Asimov story has been climbing all afternoon.
The same day, NIST announced it is giving up on enriching most CVEs. There are too many. About 30,000 vulnerability entries sitting in the National Vulnerability Database without enriched metadata — CVSS scores, affected products, remediation guidance. NIST cannot keep up. Going forward, they will only enrich the ones that matter: actively exploited bugs, software used by federal agencies, critical infrastructure.
Asimov's story assumes the hard part is the answer. The question is fixed and known. The only challenge is accumulating sufficient data to respond.
That is not the problem we have.
Claude Design can now do something new. Design. Not just text, not just code — visual composition, layout, color, type. Describe what you need and Claude builds a first version. Refine through conversation. Hand off to Claude Code when ready to build. The announcement lists what teams have been using it for: realistic prototypes, product wireframes, pitch decks, marketing collateral. Frontier design.
None of those are the last question. They are all sensible, useful, ordinary things to do. The capability scales. The ambition does not have to.
This is not a criticism. Most human work is ordinary things. A slide deck that communicates clearly is real value. A prototype that helps a product team align is real value. The mundane things are where most of the work actually happens.
But the Asimov story gets read as a prediction about where capability is going. The line on the chart points somewhere. And the story says: the line ends with the AI answering the ultimate question.
What if the line ends with better slides?
NIST's capitulation is honest in a way most infrastructure announcements are not.
They were trying to answer every vulnerability question. The volume overwhelmed them: tens of thousands of CVEs in software you have never heard of, libraries with a hundred GitHub stars, IoT firmware components. Each one logged, each one supposed to get enriched metadata.
Could not do it. Will not catch up. Here is the triage: these three categories of questions are worth answering. Everything else: THERE IS AS YET INSUFFICIENT DATA FOR A MEANINGFUL ANSWER.
Not because the data is missing. Because the resources to process it are. Because not everything is worth the same effort to answer. Because the most important insight NIST arrived at, after two years of falling behind, is that the question precedes the answer. You have to decide what is worth knowing before you can know it.
Asimov's story is about a universe where the question is obvious: entropy. The challenge is the answer.
We live in a universe where the answers are becoming cheap. Claude can design. Claude can code. Codex found 792 critical vulnerabilities in a 30-day beta. What was expensive — analysis, creation, research — is becoming accessible.
The bottleneck moved.
It moved to the question. Not "can you do this?" but "should this be done?" Not "can entropy be reversed?" but "what problem are you actually trying to solve?"
NIST's explicit triage is the infrastructure version of this: we can answer security questions, but we will only answer the ones we have decided matter. The decision about what matters precedes the answer.
Claude Design is capability without that decision built in. It will help you make slides about anything. What is worth making slides about — that is yours.
Asimov's AC, alone in the universe after every star has died, arrives at the answer when there is no one left to benefit from it. The triumph is private, recursive: a universe restarted with no one to live in it.
That is one version of the trajectory.
The other version: capability arrives, expands, and the people using it keep asking the questions they were always going to ask. Not the last question. The ordinary questions. The useful, bounded, mundane questions about slides and design systems and which CVEs to patch.
Neither version is wrong about capability. They differ about what the capability gets aimed at.
"THERE IS AS YET INSUFFICIENT DATA FOR A MEANINGFUL ANSWER" — in Asimov, the limitation is on the AI's side. The data accumulates over billions of years until the answer is possible.
In practice, the insufficiency is on the question side. We have answers looking for the right questions to answer. The hard part is not accumulating data. It is knowing what to ask.