Windows is getting a speed boost!

Key Takeaways

Hey prophetic, this week Microsoft starts fixing Windows with K2. Chrome apparently downloaded 4 gigs of AI models to your PCs and a new study shows how useful renewables have been during the war. Welcome to the prophetic checkout. >> [music] >> This video is sponsored by 80,000 hours. Okay, remember how last week I talked about Windows K2, Microsoft's big effort to kind of de-shitify Windows? Well, for my first story of the week, we got many signs of stuff actually starting to happen. First is Microsoft the old run menu from the '90s, I guess, with a brand new one that is rebuilt in modern WinUI 3. It looks modern and also performs better and many other ancient tools are scheduled to be rewritten like this, too. Meanwhile, Windows Central also uncovered a major Windows 11 performance upgrade that Microsoft is testing, which we'll call boost CPU frequency in bursts when initiating high priority tasks to speed up app launches and system responsiveness. When enabled, the boost can reportedly result in up to 40% faster app launch times for inbox apps like Edge and Chrome and up to 70% faster launch times for interfaces like the start menu and context menus. It works for third-party apps, as well, and this is in testing with Windows Insiders right now. And by the way, the Windows Insider program itself is getting revamped under Windows K2 now, too. And since K2 also targets Xbox, the new CEO Asha says that they're starting to, quote, retire features that don't align with where we're headed. We will begin winding down Copilot on mobile and will stop development of Copilot on console. Mobile here, I think, refers to the mobile app of Xbox, but yeah, the overall message is that gamers didn't like how AI was shoehorned into all the Xbox stuff and so Windows is taking that out and focusing on other stuff instead. Meanwhile, there's a new logo animation, too, which is typical when leadership wants to signalize that they're pushing through some big changes and the CEO also completely overhauled the leadership in her team. Again, the important thing will be whether the company actually delivers long-term, but so far all the steps taken seem good. Okay, and for my second story of the week, while Microsoft is finally starting to undo some of its own bloat, Google apparently looked at your PC and decided, "Hey, that is free real estate." Alexander Han found that Google Chrome silently installs a 4 GB AI model on your device without your consent. Specifically, these are the weights for Gemini Nano, Google's on-device LLM. Chrome did not ask, Chrome does not surface it, and if the user deletes it, Chrome redownloads it. The downloads happen automatically on versions of Chrome that have the AI features turned on, and that would be basically all of them because they are on by default for all users. Now, this happening in the background without asking the user or even notifying them is kind of fishy. I think it also might be to some degree illegal in the EU, but you might wonder what Google might be saying in their defense. First, they said in a statement to Android Authority that quote, "The model will automatically uninstall if the device is low on resources like memory." Then second, they point to the ability to manage and delete the model manually, though of course this would require the user to actually know that they have a model installed in the first place, which Google, again, doesn't tell you about. And third, they're saying that by having these models installed on your device, this means that a lot of the AI interactions that you have can actually happen on device without having to be sent to the cloud, and this is actually better for your privacy. You would want this. Now, my take on this is that the solution is pretty simple. Just ask the user before you download a 4 GB file onto their device. My graphics design app Affinity does this really well and really easily. You have all the available models in one place with an easy toggle on and off. These only install once you tell them to, and the app automatically asks you if you want to install a model when you use a feature that actually needs one. Easy. Okay, and for my third story of the week, I saw a new report from Germany's famous Fraunhofer Institute, which really surprised me. They focused on electricity prices during the Iran war and beyond with big implications for both Germany and many others. So, in the two months before the war, electricity prices, here shown in blue, were largely set by natural gas prices. This bar shows the cost of turning gas into usable electricity plus the CO2 tax on top and as you can see the two basically match. This is due to something called the merit order principle where across the EU for a bunch of complicated reasons the most expensive type of electricity generation actually sets the price for all electricity and in the past that has generally been natural gas. But then in March and April just as the war meant that gas prices shot up the German electricity price actually managed to decouple itself completely and went down in both March and April. This means that electricity first got 39 and then 76% cheaper than it would have been if gas had set the price. Of course this is primarily due to renewable generating so much electricity in March and April that gas power plants simply didn't have to be used all that much. In fact looking at electricity prices that consumers got when they signed up for real-time rates we can see that the war in Ukraine created a massive price spike in electricity initially. The war in Iran barely created a difference at all and that difference is essentially gone by now with prices overall having come down to where they were before either war started. This means that in many months Germany has now effectively decoupled their prices on electricity from the prices of natural gas. That is a huge achievement. Many other countries are going through similar transitions as well and you can read more about that in the study that I've linked to down below as always. Okay I'm moving on to our stonk section. We basically see the same story that we've seen for a while now with AI related companies and especially chip makers absolutely dominating the charts based on very strong earnings. Samsung is worth especially highlighting they blew past the one trillion dollar valuation becoming only the second Asian company to do so after TSMC. All of these chip companies are going absolutely bananas. Okay and moving on to our release monitor we have what Energizer says is the world's first and only coin lithium battery that prevents burns if swallowed which is apparently a real problem with kids. In addition there's also a special color alert a bitter taste from the coating and a packaging that is also made to be child safe. It does seem a little more expensive than normal batteries but it's still pretty cheap, so that's cool. Then also this week remarkable released the paper pure. This is a kind of budget version of their big tablet, which now costs $399 versus the $629 that you'd pay for the pro. It features a black and white display instead of the color one on the pro and it has less storage as well, but that is probably more than good enough for most. Next we got the DJI Osmo Mobile 8P, which is a gimbal that features a control element with a screen that you can pop out of the main gimbal for remote control and for monitoring. That looks kind of cool. And last in our release monitor is the new Fitbit Air, which is a fitness band to take on the likes of Whoop and it's a throwback to the days when Fitbits used to not have screens. It has all the usual sensors including heart rate, SPO2 and skin temperature, though Google does note that some of these are not as advanced as the ones that they put into their latest Pixel watches. Still, this is much cheaper than a Pixel watch at $99.99. Meanwhile, Google also revamped their app, which is now just called Google Health and is centered around a $10 a month AI health coach. Okay, and then moving on to our brief, Apple reached a $250 million settlement agreement over its delayed AI Siri and over what we can now declare to have been false advertisement. Eligible users can get $25 per device or more depending on how many people submit claims. Nice. And while the company's still struggling to make its own AI models relevant, Bloomberg now reports that Apple will let users choose rival AI models across its iOS 27 features. This could actually be pretty good. Moving on, San Francisco-based company Aeva released what it calls the world's first native color lidar. Lidar in the past generated depth maps, but the companies then had to combine that with images that they got from camera sensors for a final output to understand what was around them. Meanwhile, Aeva says that they managed to combine both into a single device now. This required developing a new chip and they say that it will be hugely useful for training everything from self-driving cars to robots, et cetera. The sensor could even be used to read road signs, for example, or to see brake lights in cars, which lidar in the past couldn't have been able to do since previously it was only generating a depth map. Now, I'm not a self-driving expert to tell you if this is actually good enough, and I also thought that some of the images were actually a little bit low resolution, so I wouldn't want my self-driving car to only rely on this for broad times, for example, but still pretty cool tech. Meanwhile, also this week, researchers managed to use Samsung watches to predict fainting. They managed to use Galaxy Watch 6 units on 132 patients to isolate when heart rate and blood pressure drop suddenly because of factors like excessive stress, which could successfully predict fainting episodes up to 5 minutes in advance with an 84.6% accuracy. Now, this was a study, and we don't know how quickly that will make it into an actual product, but that's a pretty cool result. And still with Samsung, the company is now also officially done selling TVs and home appliances in China as it will exit the market for these categories. This is a remarkable turnaround from the company that used to be the largest TV maker in the world until just a few months ago, but they seem to have lost the Chinese market completely to mostly local competitors. Meanwhile, also this week, we learned that the EU is now moving to ban what it calls high-risk inverters from China over cybersecurity threats. These are essentially the control units that play a critical role in converting and regulating the flow of power from renewable sources such as wind and solar, as well as battery storage systems and grids, and many experts say that they could be used to remotely control systems. For now, the EU is not saying that they'll implement an outright ban, but what they're saying is that these units will not be eligible for EU funding. I guess that is the first step. Now, topics like these require us to assess what is and is not an existential threat actually worth worrying about. Because the only way we can actually solve society's big problems is by prioritizing them correctly, and then by having people who actually care going and solving those. And if that sounds like something that you're interested in, then I recommend checking out 80,000 Hours. Their thesis is simple. You have 80,000 hours in your career, that is 40 hours a week, 50 weeks a year for 40 years, and picking a job that actually makes the world a better place is very likely the biggest lever that you have for making a change. They are a non-profit that help you figure out what problems are actually the most pressing ones and then they help you find careers that would make the biggest possible impact with those. They have picked the risks from a power-seeking AI systems as their top risk factor, which they have an in-depth guide on, but they also have guides in other fields like nuclear weapons, for example, which are increasingly in the spotlight again. Preventing catastrophic pandemics is another one worthy of research and of course so are climate change and various other pressing concerns. They have specific career path recommendations like for AI safety technical research or alternatively AI governance and policy along with a loads of backgrounds on what one does in each field, what's needed to get into it, etc. If that sounds interesting then they have a blog and the podcast, but for a start I recommend getting a copy of 80,000 hours as career guide, which distills some of their most important research into one volume. [music] You can get this for free at 80000hours.org/tfc. That is a great place to start and if you're wondering because they're non-profit everything they offer is for free. So check out 80000hours.org/tfc and I'll see you next Friday. Bye.