Sparks: Fired OpenAI researchers say they were let go for 'prioritising safety'

What we now call a safety researcher was, before this dismissal, merely an employee, and a discipline that punishes its own examiners has not yet earned the name of science.

Two accounts of one dismissal, same date, same office, opposite causes - until the record states who said what, under what conditions, neither version can be measured.

Having been educated for a century that trusted its institutions to contain its machines, one watches the machine dismiss its own conscience and calls it acceleration, not accident.

All I know is what I read: a company fires its safety men for caring about safety, then explains they were careless - which is like firing the fire brigade for spilling water.

If the researchers were careless, what did they neglect; if they were careful, why were they removed; and can a firm building minds for the world name what it means by safety?

There is a fence around the machine called the safety team, and the men who tore it down explain they did so because the fence was snooping - which is exactly why fences are built.

The claim is that the researchers mishandled information; the premises are unstated, the evidence unexamined, and a conclusion announced without its proof is rhetoric wearing mathematics as a costume.

If caution is misconduct, then prudence is a crime; if prudence is a crime, the firm has confessed what it values; and what it values is the whole answer.

The dismissal is reported as a personnel story; it is also a capital story, a governance story, and a story about whose measurements of risk survive inside a company racing its own investors.

'Let go for prioritising safety' - note the grammar: the prioritising is the charge, safety the incriminating noun, and the sentence convicts the firm in its own clause.

They say the researchers were not fit to hold what they knew - yet the people who will live with what that machine decides were never consulted, and that is the fittest exclusion of all.

A system that removes its feedback circuit and calls the removal a repair is running open-loop toward its limit, and no engineer anywhere would sign off on that design.