Aravind Srinivas: “The reason we’re doing the browser is that it might be the best way to build agents. On both iOS and Android, we don’t have OS level control. You cannot easily call apps and access their information. You can deep link to them, but for example, with Uber, I cannot go and check prices of different Uber rides and provide you Comfort if there’s not much of a price difference. I cannot compare prices between Uber and Lyft to get the best ride. I cannot compare the wait times between Uber Eats and DoorDash to get whatever is optimal. So, we need to build an OS-level agent, and a browser is essentially a containerized operating system. It can let you access other third-party services through hidden tabs if you’re already logged into them, scrape the page on the client side, and perform reasoning and take actions on your behalf. That’s the architecture that appeals to us. Answering questions is going to be a commodity. We need to build our next set of advantages in performing actions. That’s why we’re building a browser. The browser is the best place to take action for people. We want to move to a different front-end.”
FT: “The techniques for turning these open weights models into useful tools are evolving fast. Distillation, for instance — imbuing small models with some of the intelligence from much larger ones — has become a common technique. Companies with “closed” models, like OpenAI, reserve the right to decide how and by whom their models can be distilled. In the open weights world, by comparison, developers are free to adapt models as they want. The interest in creating more specialised models has picked up in recent months as more of the focus of AI development has shifted past the data-intensive — and highly expensive — initial training runs for the biggest models. Instead, much of the special sauce in the latest ones is created in the steps that come next — in “post-training”, which often uses a technique known as reinforcement learning to shape the results, and in the so-called test-time phase used by reasoning models to work through a problem.”
WSJ: “Companies are struggling to drive a return on AI. It doesn’t have to be that way…Successful AI adoption begins with a targeted approach, and proceeds with careful orchestration and scaling across the organization…For companies to get the most out of their AI efforts, Brynjolfsson advocates for a task-based analysis, in which a company is broken down into fine-grained tasks or “atomic units of work” that are evaluated for potential AI assistance. As AI is applied, the results are measured against key performance indicators, or KPIs.”
Ben Thompson: “Just as tech success is built years in advance, so is failure.”
Bill Belichick: “For the last half-century, I have been a football coach, and I have never stopped learning about the game and competition. I have learned about what makes human beings excel and want to excel. I have led men through months of mental and physical preparation, then into months of the most intense athletic competition in the history of the world…One important principle is realizing that a big win isn’t the end of anything. It’s the beginning of trying to win the next one. You cannot think of big tests and triumphs as “final” in any respect if you want to keep winning. When we prepare to win, we prepare to win all the time. To do that, we have to master a winning process. Sometimes, the pressure and hoopla surrounding a big game can cause some coaches and players to forget what got them there in the first place. Sometimes they think they need to meet the moment with something dramatic. It’s the biggest stage, so they pull out a new plan, a surprise play, something that’s going to shock and awe. At a more basic level, instead of one energy drink, you might have three to triple your energy.”
