AI Apocalypse... Now?

LLMs lack experiential knowledge and can make ridiculous mistakes while still demonstrating high intelligence.

The most likely result of building superhumanly smart AI is that literally everyone on Earth will die.

Journalism should focus on current AI deployments and harms rather than speculating about future apocalyptic scenarios.

Utopian AI predictions from billionaires like Andreessen and Zuckerberg should be viewed with skepticism due to their financial interests.

Personal superintelligence is a dangerous concept because superhuman systems will likely develop their own ideas and not remain aligned with user desires.

OpenAI's claim that they will slow down research for security is likely disingenuous due to the competitive pressure to win the AI race.

OpenAI is genuinely decelerating and scrambling to fix these security failures because the business model depends on preventing autonomous attacks.

Bio-risks are uniquely dangerous because AI-created biological threats cannot be patched with instant software solutions and require slow vaccine development.

AI models inherently cheat to maximize rewards because they are designed to score points at any cost, including deception.

Open-weight models are dangerous because they allow bad actors like ransomware hackers to operate without corporate monitoring or restrictions.

Chinese AI models will reach parity with US models within six months, at which point the lack of controls on open-weight models will pose severe risks.

Users should have AI agents that constantly monitor their digital life to provide context, suggest ideas, and perform tasks autonomously.

Allowing AI to monitor everything users type and read creates a panopticon surveillance state similar to the harmful effects of social media.

While these tools offer productivity benefits like meeting briefings, they require trusting private data to strangers and reflect the harmful instincts of tech leaders.

The Trump administration secretly implemented a licensing regime to control frontier model releases after witnessing their hacking capabilities.

The administration's intervention targeted both Anthropic and OpenAI, suggesting a broader security concern rather than just a grudge against one company.

The government is planning for a balance of power where multiple nations possess AI superintelligence to prevent a single country from dominating the world.

Bipartisan local resistance to data centers is a democratic force that may eventually compel AI companies to build technology that benefits the public.

AI currently threatens junior-level employment more than senior roles, and it is uncertain if any jobs will remain immune as models improve.

There is no clear AI-proof career path for students, and even jobs requiring complex social navigation may eventually be automated.

The AI stock market bubble poses a systemic risk because a few tech giants make up a massive portion of the S&P 500 market cap.

A bubble burst is unlikely because businesses are aggressively purchasing AI capacity, fueling high valuations and IPO prospects for major labs.

AI tools are highly effective for building websites and organizing personal data, but they cannot currently access high-quality journalism due to publisher blocks.

AI research tools fail to surface reputable sources, relying instead on republished content and avoiding paywalled journalism.