Skip to content
LessWrong· loops·Indexed by this site · Originally published· Opinion →· Its place in the scan record →

The author worries that LLMs accelerating hardware R&D makes cheap ASI easier to achieve

I'm worried about LLMs accelerating hardware for cheap ASI

AI summary

The author worries that LLMs accelerating hardware R&D will pave the way for building cheap ASI through a new paradigm, and that such ASI will most likely be misaligned. By their estimate, with suitable hardware and algorithms, the energy cost of new-paradigm ASI could be below $10k; the bottleneck may be hardware, and LLMs are becoming good at hardware design and FPGA programming. The author suggests AI labs avoid making models too good at hardware design and implement safeguards preventing their use to develop new hardware.

Read originallesswrong.com