Jeffrey Ladish-AI Cyberwarfare, Compute Monitoring

[@TheInsideView]

OUTLINE 00:11 The Future of Automated Cyber Warfare and Network Exploitation 03:30 Evolution of AI in Cybersecurity: From Source Code to Remote Exploits 07:56 Augmenting Human Abilities with AI in Cybersecurity and the Path to AGI 12:47 Enhancing AI Capabilities for Complex Problem Solving and Tool Integration 15:57 AI Takeover Scenarios: Hacking and Covert Operations 17:42 AI Governance and Compute Regulation, Monitoring 20:23 Debating the Realism of AI Self-Improvement Through Covert Compute Acquisition 24:36 Managing AI Autonomy and Control: Lessons from WannaCry Ransomware Incident 26:36 Focusing Compute Monitoring on Specific AI Architectures for Cybersecurity Management 29:41 Strategies for Monitoring AI: Distinguishing Between Lab Activities and Unintended AI Behaviors

[@ikotsus2448]

I don't understand slow takeoff. If my feeble human mind operated tirelessly at an orders of magnitude faster timescale with all the knowledge of the internet I would consider myself superintelligent. Am I wrong?

[@RougherFluffer]

Not your fault. They are ill defined and counterintuitive concepts. Current AI is indeed superhuman in many ways, but not as generally capable as an average human, particularly in physical domains. By the time we have an AI that can match any human in any conceivable domain, it will be superhuman in a wide range of subject matter. Superintelligence is to the whole of humanity what AGI is to a single human. A Superintelligence will be more intelligent, more capable, more productive, than the entierly of modern humanity. As for slow vs fast takeoff; personally I feel we are already in a human driven, AI assisted, slow takeoff scenario, but that could escalate to a faster takeoff with breakthroughs in AI self-improvement. A slow takeoff is the continued exponential trend we've witnessed consistently for decades in computing capability gain. A fast takeoff might look closer to hyperbolic improvement.

[@ikotsus2448]

@@RougherFluffer "Superintelligence is to the whole of humanity what AGI is to a single human." This helps a lot. But still, as humanity can be bottlenecked by a few people, meaning that 10000 scientists/engineers may not do a task 10000 times faster than 1, I am still pessimistic on this one. One could argue that some objectives could be bottlenecked by interacting and experimenting on the real world, but an abundance of consideration could find paths that do not require these.