OpenAI researcher says AI now runs whole training experiments
Aidan McLaughlin, who works on at OpenAI, said it has become routine for him to let “5.6 e2e” carry out an entire run. Nathan Calvin interpreted this as AI taking a larger role in building better . He also cited separate data claiming that OpenAI’s internal computing use grew 100 times in six months, arguing that the cycles between experiments and new training are getting shorter.
The exact identity of “5.6 e2e,” the amount of human involvement, and any measured improvement were not disclosed. This is therefore an early anecdote, not proof that has been achieved.
Key points
- An OpenAI er says “5.6 e2e” routinely handles a full run.
- OpenAI’s internal computing use was claimed to have risen 100-fold in six months.
- The claim suggests shorter cycles between AI experiments and further training.
- No details show how much human supervision remains or how well the runs perform.
- There is no direct evidence yet of .