Release gaps: from 877 days down to the 20-day range

A summary of a conversation that uses the numbers on model turnover cycles to explain why the call to slow down came now, of all times. In one model family, the gap between top models shrank from 877 days to 336 and then to 56; another family has come down to 63, 23 and 20 days, presented as a state in which a new model arrives almost once a month. Recursive self-improvement, models building the next models, is singled out as the cause, and an internal tally from one research organization, in which agents' running time was 3.1 times human working hours, is also cited. Of the 3 proposed stages, only stationing an external evaluation team on site is settled, and the rest work only if everyone slows down together, drawing the criticism that it is a one-sided restriction with no technology sharing. It concludes that if IPO preparations and a shortage of compute are the backdrop, the one that gains from this setup is Google, which had been falling behind in the race; but the sources of the figures are not disclosed, so they are hard to cite as is.

This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.