Washington Hones in on AI Safety, What Are AI Doomers Proposing | Diet TBPN
TBPN hosts John Coogan and Jordi Hays break down the AI 2040 proposal, a concrete regulatory plan to delay superintelligence from 2028 to 2040 without banning AI entirely.
The gist
-
The AI 2040 proposal allows inference on existing models but pauses new frontier training runs, targeting 2040 instead of 2028 for superintelligence.
-
Compute caps are the main enforcement tool: data centers with 10,000 or more H100-equivalent GPUs would need independent auditors to verify workloads.
-
Physical security measures like Faraday cages and 1 megabit per second bandwidth caps are borrowed from nuclear non-proliferation frameworks to slow capability gains.
-
Aggressive regulation risks pushing AI research underground and enabling regulatory capture where only around 10 large companies survive.
What they recommend · 1 pick
"Computer use in iRacing is something I'm going to experiment with this week."
Also mentioned
Related episodes
Episodes covering similar topics