AI Safety Is In More Trouble Than People Realize. Experts Attack Each Other in Viral AI Debate
Tom Bilyeu breaks down a viral expert debate on AI safety, explaining why disagreement stems from different definitions of AI and what the evidence actually says about runaway superintelligence.
The gist
-
Experts talk past each other because they conflate today's LLMs with hypothetical self-improving superintelligence, making useful policy nearly impossible.
-
AI improvement rates are plateauing at around 9% per generation, below the 15% threshold needed for a self-sustaining intelligence explosion.
-
Bilyeu says control is already achievable through intentional design, adversarial defenses, and monitoring, similar to how humans manage systems like AlphaFold.
-
Rather than fearing a sci-fi takeover, he recommends setting specific behavioral thresholds that trigger automatic kill switches before deployment.
Tom, This Week in Startups, Greg and Ali agree on 20 things
See what they all use, in their own words. Paid ads are left out.
See what they all swear by Or pick your own peopleAlso mentioned
10:05
ChatGPT
· "It would appear to me that the new ChatGPT-6 model, the Fable 5.1 model, is arguably smarter than 99.999%..."
10:58
AlphaFold
· "If you think about AlphaFold, the AI that does that is truly far superior to humans in what we're calling..."
29:14
Hugging Face
· "Unplug that data center and the agents stop. So this is exactly what happened with the Hugging Face..."
Ads in this episode
Pipedrive, Quo. They're paid, so we keep them out of the picks above.
See what Tom recommends without being paid →
◎
Related episodes
Episodes covering similar topics
More from Tom Bilyeu
The Brutal Truth About Why Hard Work Feels Pointless
Oct 15, 2026
How One Shane Gillis Joke Split the Internet in Half
Oct 14, 2026