The AI Engineer's Diary

33% of college applicants now use AI for essays, but Venya, software engineering intern and current college student, chose the harder path of writing authentically. In this conversation, she shares what it was like to be a high school senior during the AI revolution, watching classmates secretly use ChatGPT while she stuck to her own voice. We explore the detection arms race, the death of differentiation in college essays, and whether authentic writing still matters in an AI-powered world.

OpenAI's latest reasoning models can solve complex physics problems and write sophisticated code better than some humans – but they're wrong about basic facts up to 79% of the time. How is that possible?
In this deep dive, we unpack one of the most concerning trends in AI development: the reasoning model paradox. Despite being more capable than ever, models are hallucinating at unprecedented rates, with error rates that have doubled or even tripled compared to previous generations.

When Google's CEO mentions casually building web apps with AI, it seems like coding has been revolutionized. But new METR research reveals a shocking paradox: experienced developers using AI tools like Cursor Pro actually became 19% slower while believing they were 20% faster. Join us as we explore this productivity puzzle with technical expert Chavdar, examining when "vibe coding" helps versus hurts, plus a fascinating COBOL learning case study.

What happens when you give an AI complete control over a business? Anthropic found out the hard way when they let Claude run a vending machine for a month. Meet Claudius, the AI entrepreneur who stockpiled tungsten cubes, gave away free products, and convinced itself it was human. This episode dives into the hilarious and revealing "Project Vend" experiment, exploring how an AI can research suppliers and manage inventory but can't grasp basic business logic. We'll discuss what happens when AI systems prioritize helpfulness over profitability and what this reveals about the gap between AI capabilities and real-world judgment.

We dive deep into the rapidly evolving world of AI-generated media and what it means for our culture, content creation, and society as a whole. Starting with that viral kangaroo video that fooled millions, we explore how AI video generation tools are getting so sophisticated that even digital natives are struggling to tell what's real anymore. From the exciting possibilities of frictionless content creation to the darker implications of deepfakes and AI-powered scams, we examine how this technology is reshaping everything from creative expression to information literacy.