Home › Hard problems › Data and computer systems

Data and computer systems

Proving AI-written code is correct

Solo: strongOne person can run a real test of this in about 8 weeks.

AI writes codehundreds of linesin secondsTestscatch some bugs,miss othersShippedwith the bugsnobody sawfaster than anyone can reada person reads slowlyA formal proofLean or Dafny can show the code is right,but they are hard to usethe sure way, rarely taken
AI writes code faster than people can review it. Tests catch only some bugs, and the tools that can prove code right are hard to use.

Why it is still open

AI writes code faster than people can review it. Tests catch some bugs; proof tools (Lean, Dafny) are hard to use.

An experiment one person can run in 8 weeks

Have an LLM write small functions plus a formal proof in Dafny. Measure how often the proof passes and the code is truly right.

Verdict: Solo: strong

Free tools, and results are pass or fail.

The solo test behind the verdict

  1. Can it be measured? There is a number that shows you did better, such as an error rate, a catch rate or a cost per answer.
  2. Is it cheap to try? It runs on a laptop, free open models and open data, or under about ₹50,000 of rented GPU time.
  3. Is there a public yardstick? A benchmark or dataset others use, so your result can be compared and believed.
  4. Would a small win get noticed? A lab, a funder or a customer would care about a 10% improvement, not only a full solution.

Four yeses means Solo: strong. Two or three means Solo: possible. Fewer means Follow, don't build.

PreviousKeeping thousands of GPUs busyNextSame input, different LLM answer

All 79 hard problems · Working on this, or know of a result we missed? basant@cleandataengineering.ai