Fetching from the wire…
Public story · 2026-08-17 · high
The GAMESS-based harness ran 2,200 fault injections across twelve sites and turned up parallel deadlocks and false convergence.
Why now: Covered in the 2026-08-17 briefing.
A GAMESS fault-injection harness fired identical faults at the original and its LLM-modernized version, per arXiv 2608.14527. It ran across twelve sites and more than 2,200 runs. The question wasn't whether the rewrite gets the right answer on clean input. It was whether the rewrite fails the same way the original does under stress.
The original and the rewrite agreed on all 200 paired fault injections. The campaign also turned up phase-dependent parallel deadlocks and false convergence under reduced precision, per the paper. Nominal testing, the kind most modernization projects rely on, misses both.
That's the gap most AI code migration reviews miss. A rewrite can match the original on every clean run and still diverge under stress. A race condition, a precision drop, or a timing edge case can slip past a passing test suite.
The paper doesn't say how common this gap is outside scientific computing. GAMESS is long-running, precision-sensitive, and parallel, not typical enterprise software. But differential fault injection, not nominal testing, is what actually shows whether the rewrite still fails like the one it replaced.
Each link below shares sources, entities, or timing with this story.
Simon Willison released LLM / Same source domain / Shared topic / Tension
Linked by a graph relationship (Simon Willison released LLM); reported by the same outlet (arxiv.org); overlapping topics (apply, injection).
Simon Willison released LLM / Shared entity: LLM / Earlier coverage / Tension
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-07-27.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-08-16.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-19.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-18.
Simon Willison released LLM / Shared entity: LLM / Earlier coverage
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-07-31.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-19.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-08-12.