May 2026

Training large language models requires accurate feedback signals, but traditional reinforcement learning (RL) often struggles with reward signal reliability. The quality of these signals directly influences how models learn and make decisions. However, creating robust feedback mechanisms can be complex and error prone. Real-world training scenarios often introduce hidden biases, unintended incentives, and ambiguous success criteria that can derail the learning process, leading to models that behave unpredictably or fail to meet desired objectives. In this post, you […]

Ver mais

I trained an AI to speedrun Super Mario Bros using Reinforcement Learning — after more than 6 million deaths 😅 The agent starts completely clueless: running into the first Goomba falling into pits getting stuck against pipes Over time, it slowly learns: movement timing enemy avoidance jump precision speed optimization What’s interesting is that some “speedrunner-like” behaviors emerged naturally during training: maintaining momentum minimizing hesitation optimizing jump timing The training was done using a custom RL setup with […]

Ver mais

Wearables have really come full circle. The early Fitbits didn’t have screens, but the move to smartwatches put a screen on everyone’s wrist. Now, devices like Whoop and Hume are designed as data trackers first and foremost without so much as a clock. Google’s newest wearable jumps on that trend: The Fitbit Air doesn’t have a screen, but it does have a suite of health sensors that pipe data into the new Google Health app. And if you […]

Ver mais

Last fall, we featured an extensive interview with Petter Törnberg of the University of Amsterdam, who studies the underlying mechanisms of social media that give rise to its worst aspects: the partisan echo chambers, the concentration of influence among a small group of elite users (attention inequality), and the amplification of the most extreme divisive voices. He wasn’t optimistic about social media’s future. Törnberg’s research showed that, while numerous platform-level intervention strategies have been proposed to combat these […]

Ver mais

Image created by the author using Figma If you’ve worked with tabular data, you’ve likely noticed something: No matter how advanced deep learning becomes, tree-based models often outperform everything else. From credit risk prediction to medical decision support, models like XGBoost, LightGBM, and CatBoost dominate real-world machine learning tasks. But why are they so powerful? And how did we get from a simple decision tree to these highly optimized algorithms? This article breaks it down from first principles — in a way […]

Ver mais

Elon Musk tried to hire OpenAI’s founding team, including Sam Altman, to lead a new AI lab within Tesla in 2018, as the AI start-up’s leaders grappled over who should control the company and its direction. Musk, a co-founder of the AI group, proposed bringing Altman, Greg Brockman, and Ilya Sutskever to his carmaker, appointing Altman to the board or making OpenAI a Tesla subsidiary, according to evidence in a high-stakes trial between the billionaire and the ChatGPT […]

Ver mais

The productivity gains are real. The compound debt is realer. And almost no team is measuring the right thing. Photo by Bernd 📷 Dittrich on Unsplash Let me say the quiet part out loud: vibe-coding is not the villain in this story. The critics who call it “reckless” are wrong. The evangelists who call it “the future of engineering” are also wrong. And if your team is using it at scale without a framework, you are quietly building a time […]

Ver mais

We’re in the midst of a fundamental shift in how software gets built and used. AI agents are moving beyond assistants that wait for instructions. They call APIs, access MCP servers, coordinate with other agents, and complete complex multi-step tasks on behalf of users. As agents take on increasingly diverse tasks, the ecosystem around them is expanding just as fast to meet that demand. Looking further ahead, services, tools, and content must be designed for humans and agents. […]

Ver mais