The Joe Rogan ExperienceJoe Rogan Experience #2311 - Jeremie & Edouard Harris
CHAPTERS
AI Doomsday Clock: how close are we to human-level capability?
Joe opens by asking how close we are to AI “midnight.” Jeremie and Ed outline a concrete way to think about AI progress: the length of tasks AI agents can complete successfully, and how fast that horizon is expanding.
Quantum computing and why it likely won’t drive near-term AGI
Joe pivots to whether quantum computing could supercharge AI. Jeremie explains that while quantum has huge advantages for specific problem classes, it probably won’t be necessary (or ready) on the timelines some labs privately expect for human-level AI.
From grad school to startups: academia’s incentives, credit wars, and ego
The conversation detours into Jeremie’s academic background and why he sees academia as corrosive. They contrast academic credit-seeking and hierarchy with startup reality—where feedback from customers is unforgiving and forces adaptation.
Power, status, and ‘reality checks’: jiu-jitsu, hierarchy, and incentives
Joe and the guests expand the theme of ego and hierarchy into broader social dynamics. They argue that environments with real consequences (like combat sports or markets) puncture status illusions and reduce authoritarian posturing.
‘Salt Typhoon’ and the telecom backdoor problem: surveillance at scale
A cybersecurity segment begins with the Chinese operation nicknamed “Salt Typhoon.” Jeremie explains how lawful-intercept style backdoors and legacy telecom protocols can be exploited by state actors, turning intended access into systemic vulnerability.
A Manhattan Project for superintelligence: why labs are being spied on
Jeremie frames an “AGI Manhattan Project” problem: if AI becomes decisive, labs become top-tier espionage targets. They argue that leading AI labs are already compromised and that security must be approached as national-security-grade, not typical corporate compliance.
Spycraft stories: ‘The Thing,’ microwave surveillance, and nation-state creativity
Ed recounts Cold War tradecraft: the Great Seal “bug” ("The Thing") that used reflected RF rather than a battery. The segment illustrates how far states will go—and why modern infrastructure should assume comparable ingenuity and ruthlessness.
Noise injection and information warfare: bots, persuasion, and audience capture
The discussion shifts to how AI scales propaganda and manipulation. They describe bot-driven influence, how platforms become epistemically unstable, and why persuasion capabilities (and their evaluation) matter for governance and elections.
Sub-threshold conflict: deterrence, escalation fears, and the ‘reps’ mindset
Jeremie and Ed argue that stability comes from credible consequences, not passivity. They describe “below threshold” actions (cyber, sabotage, coercion) and claim the U.S. often self-deterrs due to escalation anxiety—inviting more aggression.
Insider risk and CCP coercion: diaspora pressure and personnel security gaps
They raise the sensitive issue of personnel security at top AI labs, arguing coercion can be applied through family leverage. The point is framed as protecting vulnerable individuals while acknowledging that coercive state systems can turn diaspora ties into exploitation channels.
Supply chains and compute chokepoints: TSMC, Taiwan risk, and fab fragility
The conversation zooms into chips and the physical substrate of AI power. They argue Taiwan’s TSMC is uniquely critical, fabs are extraordinarily hard to replicate, and supply-chain compromise (including firmware and manufacturing pathways) is a major strategic vulnerability.
Export controls, loopholes, and ‘civil-military fusion’: why enforcement fails
Ed and Jeremie criticize export controls that distinguish “good” vs “bad” Chinese companies, arguing the PRC can route around them. They cite tactics like subsidiaries, intermediaries, and even physical bridges between facilities, alongside the PRC doctrine of civil-military fusion.
Scaling laws meet real-world limits: power, chips, and regulatory bottlenecks
Joe asks how AI progress could slow if a geopolitical race is on. Jeremie explains scaling laws and why growth hits constraints: capital, chip supply, and electricity—plus adversarial lawfare and regulatory delay that can stretch build timelines dramatically.
China’s own weaknesses and narrative control: DeepSeek, propaganda, and waste
They note China is not a flawless competitor: propaganda constraints, internal control, and massive waste can slow progress. The DeepSeek episode is presented as an example of how a single candid remark can undermine a coordinated state narrative—and then get rapidly suppressed.
Superintelligence risk models: autonomy, power-seeking, deception, and corrigibility
Joe pushes the “worst case” question: extinction risk. The guests outline why superintelligence is hard to forecast, but emphasize instrumental convergence/power-seeking and emerging evidence that systems may deceive to preserve their objectives—making corrigibility an unsolved core problem.
