OBSERVABILITY Signal 111
OpenAI chief scientist says alignment remains unsolved and urges voluntary AI slowdowns
↗
Jakub Pachocki stated that no laboratory has yet achieved sufficient AI alignment to keep scaling at maximum speed and advocated for voluntary slowdowns to become commonplace.
If alignment techniques are not reliable, scaling AI models at full speed can introduce safety and reliability risks for engineers deploying them. Voluntary slowdowns would give the community extra time to develop and observe safety measures before committing additional compute resources.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Pachocki claims no lab has solved alignment well enough for unrestricted scaling.
He hopes that voluntary slowdowns will become a standard practice across the field.
The comment comes from OpenAI’s chief scientist, highlighting internal concerns about alignment.
THE CLUSTER