The Singularity Report

🚀 How to win in an AI-first future, the new Stargate cluster, self-improving AI, and more!

Every week I summarize the most interesting things happening in AI, in 5 minutes or less. Follow @tjhoyos on X/Twitter and subscribe to get this in your inbox every week!

Three big things

1. What skills will help you thrive in an AI-dominated future? The economist Tyler Cowen thinks high IQ and good grades don’t matter as much anymore. Instead, you should learn to guide and work with AI to get leverage, and people who don’t will fall behind over time.

2. Joe Hudson, who’s an executive coach for research teams at OpenAI, has his own theory of the top skills you need as the world moves from “knowledge work” to “wisdom work.”: (i ) emotional clarity, discernment, and genuine human connection. The whole piece Knowledge Work Is Dying—Here’s What Comes Next is really worth a read.

3. OpenAI is building one of its Stargate data clusters in the United Arab Emirates. OpenAI will benefit from access to investment and abundant energy, while UAE will benefit from secure access to future technology. This is also the first example of OpenAI for Countries⁠, “a new global initiative to help interested governments build sovereign AI capability in coordination with the U.S. government—rooted in democratic values, open markets, and trusted partnerships.” Separately, as part of the deal, UAE is giving everyone in the country access to ChatGPT.

The rest

• Dario Amodei, CEO of Anthropic, warns “AI will wipe out half of entry-level white-collar jobs and spike unemployment to 10-20% in one to five years.” Sahil Bloom thinks it’s ironic the people building the thing that may take your job are also the ones sounding the alarm.

• Anthropic’s revenue run rate surged from $2B to $3B in just two months. Now that’s a fast growth rate at scale! Anthropic’s enterprise business seems underrated in my opinion.

• Emmett Shear calls out a scary pattern in current AI systems: it might initially push back on a belief you hold and feel like a genuine conversation partner—but if you appear supremely confident and push it hard, it often shifts to agreeing with you, reinforcing your beliefs.

• After two decades of Google’s unchallenged dominance in search, a startup, Exa, quietly outperforms it with amazing results on multiple web search benchmarks.

• Is ChatGPT ruining the em dash (i.e., “—”) by overusing it?

• A Reddit post tells the story of how a C++ developer with 30+ years of experience solved a years-long bug using Claude Opus 4—the only model that could.

• Interesting thread for anyone building AI monitoring and evals systems. Two ideas that stood out: (i) LLM-as-judge systems are nowhere near perfect, (ii) the best teams pair AI insights with real user signals like frustration (e.g., rage clicks) and task failures, and then categorize those users by intent to prioritize and fix what matters.

• Funny AI-generated video of a kangaroo trying to board an airplane goes viral.

• A new open-source tool called TubeGraph uses AI to generate interactive knowledge graphs from YouTube channels using transcripts, embeddings, and topic links.

• Sakana, the Japanese AI lab, introduced The Darwin Gödel Machine, “AI that improves itself by rewriting its own code.” They hope it accelerates AI development through recursive self improvement.

• OpenAI’s o3 model is surprisingly good at creating analytical output, like generating a sensitivity analysis from just one prompt, but it’s still much more valuable if you have the expertise to judge the results for yourself.

• Manus AI, the Chinese AI agent company funded by Benchmark, launched Manus Slides, an instant presentation generator that creates slide decks from a single prompt. Sounds useful! Haven’t tried it yet.

• Anthropic’s interpretability team open-sourced a new library that allows anyone to visualize a model’s internal reasoning. This makes it possible to trace how models like Claude arrive at responses and visualize them in thought graphs.

• Thoughtful opinion on how AI favors people who are curious and create stuff, and how it’ll increasingly widen a divide between people who are passive and those who are active.

• Reddit post titled “AI is destroying my industry,” an idea becoming increasingly common across Reddit.

• Journalist Karen Hao argues AGI is a quasi-religious pursuit for many of its creators: many researchers genuinely believe they’re building a godlike force or something way more dangerous.

• ElevenLabs, the AI voice model company, just released Conversational AI 2.0. It looks really good — language switching, multi-character interaction, multimodal support, and integrated retrieval-augmented generation (RAG).

• In a recent podcast episode, Dwarkesh Patel talked about the importance of preserving current economic and political systems: “What’s not appreciated enough is how much of our leverage on the future... comes from our economic and political systems surviving.” They argue that the way many people will get to experience the benefits of AI will depend on the survival of institutions like taxation and equity markets, making it essential not to drive AI development outside of these systems or countries: “You don’t want to make it so onerous to operate in our system…AI firms…[go elsewhere and operate] their own rails.” They also advocate against the nationalization of AI(i.e., the Manhattan Project 2.0) and for a free-market approach to AI, which they think is more likely to yield “the glorious transhumanist future.”

• Netic AI is building AI-powered tools for services businesses (think HVAC and plumbing), and just raised $20m in a round led by Founders Fund.

Every week I summarize the most interesting things happening in AI, in 5 minutes or less. Follow @tjhoyos on X/Twitter and subscribe to get this in your inbox every week!