27/3/2025
With progress on large language models (LLMs) stalling, techbros in the industry have had to come up with new ways to signal progress and keep billions of investors’ dollars flowing while “artificial general intelligence” (sic) remains nowhere in sight.
This led to the emergence of nonsense such as the new “magic” version of ChatGPT that’s supposedly great at “creative writing,” autonomous “agents,” and more models that can “think” or “reason.” (All in quotes because these simulations are, at best, mediocre and often non-functional.)
Amid the parade of new applications for generative AI, “vibe coding” emerged — a term coined in February by Andrej Karpathy, co-founder of OpenAI.
In broad terms, vibe coding is a complete abstraction of software development. Instead of writing… code, the developer writes prompts in natural language to an AI, describing the software it hopes to achieve. The AI then spits out code which, if it doesn’t meet expectations, is reworked in the same way: with more natural language instructions given to the AI. In this setup, the developer essentially becomes a guesser. In the end — and with some luck — the session wraps up with a working application.
Programming is a powerful skill, even outside contexts like startups and world-changing ideas.
Continue reading »
7/3/2025
Does your phone run Android? If so, you might have noticed a new app called SafetyCore. Announced by Google in October 2024, the company has been rolling it out recently (at least here, in Brazil).
SafetyCore is designed for devices running Android 9 or later, takes up about 2 GB of storage, and according to Google “provides common infrastructure that apps can use to protect users from unwanted content.” The documentation also notes that “the classification of content runs exclusively on your device and the results aren’t shared with Google.”
Almost no one reads these docs or even warnings, alerts of an app. What definitely catches your eye is a new icon among your apps that seems to appear overnight. Is SafetyCore something to worry about?
It’s published on the Play Store and, like any other app over there, it gets ratings and comments. Its average rating is 3.5, with the highest (5) receiving the most votes overall and the lowest (1) not being insignificant in number. One standout negative comment sums up the problem (in Portuguese, here translated):
“The app installed on its own and when I tried to open it, it just showed the app info. I’m not sure if it’s legitimate or not, and that worries me and many other consumers, breaking our trust in the security of the operating system. A quick question: if the app is legitimate, is it supposed to have an icon and do something, or is it supposed to be hidden? Like, for example, Google Play Services.”
If I had to sum the issue up even more, I’d say that the way SafetyCore was released is an example of a lack of transparency and a disregard for user autonomy. No matter how good Google’s intentions may be — which, judging by the history of big tech, is far from a guarantee — this isn’t the right approach.
Apple isn’t off the hook either. At the end of December 2024, someone noticed a new option in Apple Photos: “Enhanced Visual Search,” on by default. It identifies the location of photos even when there are no geo location in the metadata, by recognizing landmarks in the images. The documentation explains that it “works without sending your photos or videos to Apple and without Apple learning about the information in those photos or videos.”
In both cases, trust is lost when features like these are enabled quietly, without giving users the option to opt-out — a practice that remind what malicious parties would adopt if their intention were installing malware on millions, even billions, of devices.
What’s frustrating is that these are promising features that seem like good ideas. Scanning for unwanted or malicious content on your device without sending data to Google’s cloud? A real step forward, if true. Improving photo search without handing over your images to a big tech company? Sounds great.
However, note that both promises are extremely difficult to verify since all the code is closed source. That alone is a huge red flag. Enabling these features by default without any notice only worsens the situation.
27/2/2025
I’ve been using the my work as a tech writer as an excuse to dive into some unproductive obsessions. And when I say “obsessions,” I mean it in a pretty serious, almost clinical way.
It’s not just my work’s fault, I think. My “information diet” — reading blogs who obsess over details like which app to use for this or that, watching YouTube channels that scrutinize phone models in such detail that a normal person wouldn’t notice the difference from the last five iterations etc. — pretty much created an alternative reality that initially seemed appealing but eventually turned suffocating.
I spend an unhealthy amount of time thinking about where to host my files and emails, looking into FOSS alternatives to apps that I already have and that work perfectly fine for me, keeping up with and testing new apps that promise to be better than the ones I currently use and enjoy.
The biggest and most recent manifestation of this obsession was trading my trusted iPhone SE for an Android phone.
Switching phones isn’t just… switching phones. Even though I’m getting along fine with Android, the switch has led to even more time spent on crazy excursions in search of things that, strictly speaking, I don’t really need. It’s fun, but it’s a bottomless time sink (which I think is a waste).
In a way, indulging in this feels like eating fast food: enjoyable in the moment, yet leaving you with indigestion later and proving dangerous in the long run.
To make matters worse, the Android phone I bought is huge and heavy. Having it in my pocket is annoying, my hands ache when I try to type even a slightly lengthy message, and there’s no comfortable way to hold it for reading.
I guess I needed to make that impulse decision to try and curb the obsession. It’s kind of like a drunk taking one last swig before committing to sobriety, you know? I think it worked, because I’m having a major hangover after having a blast setting Android up exactly the way I wanted. Even so, I’m going to abort the migration — not so much because of Android, which I actually found pretty cool, but more because I miss the human hands-compatible size of the iPhone.
All of this sounds pathetic, almost comical — or tragicomic, really; it affects both me and my work. If my goal in writing about tech is to achieve a healthy relationship with technology as much as possible, I need to take a step back and rethink some things.
16/2/2025
The launch of Tapestry in early February has solidified a new category of apps — ones that attempt to create a unified timeline from different sources that, by their very nature, are like oil and water.
Tapestry joins a handful of other recent apps1 — Feeeed, the new Reeder, and Flipboard’s Surf — in tackling the main issue of decentralized social platforms, which is… well, decentralization itself.
The idea is pretty cool: it doesn’t matter if the people you follow are on Bluesky or Mastodon. With one of these timeline apps, you can keep up with them all at once on a single interface. It’s like centralizing the new social internet, but it’s done at the individual level rather than the platform level.
Continue reading »
6/2/2025
From the obvious to the outrageous, the list of concerns about artificial intelligence has grown long since late 2022, when ChatGPT took the title of “technology of the future” from the metaverse or NFTs.
I’ve been thinking a lot about one of these concerns: the excessive use of energy and water needed to satisfy the insatiable thirst of big techs and startups for more money.
What is the environmental cost of outsourcing thankless tasks to ChatGPT, like writing reports that no one reads or generating a happy birthday image for that aunt you haven’t spoken to in six years, in the family group chat?
Perhaps the most popular metric for this dilemma of the 2020s is the 500 ml water bottle for every ~50 questions to ChatGPT.
Is that a lot? A little? Is the trade-off worth it? As with everything in life, the answer is: “it depends.”
Any data point, when analyzed in isolation, sounds absurd. Context is necessary. And the more context, the better. In this case, we are particularly interested in comparisons to other everyday activities we do online, which also rely on large data centers, and the cost-benefit of adding fuel (even just a few drops) to global warming in exchange for artificial verbosity and cheesy images.
***
There are estimates that the AI race will double the tech sector’s share of global energy consumption by 2026. Which… is quite a lot, considering that the sector consumed between 1–1.3% of the total back in 2022.
What can you or I do if we find this increase wasteful? To be honest, I think nothing. For us mere mortals, AI is inevitable. Like it or not, CEOs of major companies — who have the final say on the apps and websites we use daily, and how we use them — are convinced that this technology is revolutionary.
Not just convinced: they are all in, hopeful that, like any truly revolutionary technology, AI holds substantial rewards for those who master it, even if it incurs “externalities,” including frying the planet.
The inevitability is already being felt. Google and Microsoft have integrated generative AI into their productivity apps. Overnight, millions of people and businesses relying on Microsoft 365 (formerly Office) and Google Workspace gained AI assistants they didn’t ask for, which, of course, came with a considerable price increase.
Samsung has given up on improving its smartphones — even slightly downgrading the Galaxy S25 Ultra — to focus its efforts on at least questionable AI features as highlights, in a bold attempt to convince people to upgrade their devices.
On the other side of the mobile duopoly, Apple, which refused to say “AI” and called it “machine learning” until 2022 (before ChatGPT), has embraced AI, coined an infamous pun to refer to its take on it (“Apple Intelligence”), and decided to enable it by default in iOS 18.3.
This is the same AI that, until this update, was generating fake news by “summarizing” notifications from news apps. Apple’s solution? Remove news apps from the AI summary. Apple still sees you as a selfish asshole; that hasn’t changed.
***
Given that it’s inevitable, does it make sense to oppose the use of AI? Or rather: is it possible to stop using AI-based resources, to vote with our wallets?
For the more obvious cases, perhaps yes, and it might even be desirable since they can be tacky and/or nearly useless. An AI startup backed by billions from Amazon and Google, Anthropic agrees.
The problem is that, in a broader context, we are all already using “AI” to a large extent. From searching for images in Apple/Google Photos to more innocuous routines like spell check on our phones or automatic translations, everything relies on concepts and techniques that, depending on the definition (if broader), fall under the umbrella of artificial intelligence. And they offer more tangible or justifiable benefits than just removing ugly people from the background of a photo.
And it’s this perception that leads me to a fatigue that precedes the discussion itself, because it feels futile. The infamous 500 ml water bottle that ChatGPT consumes is the new “turn off the lights to save energy,” the “take 30-second showers to avoid running out of water in the world,” meaning it doesn’t really matter whether we use ChatGPT or not because:
- The damage is already done, and it won’t be one person or a million people trying to ignore AIs (or the obvious parts of them) that will change the course set by the industry; and
- Even if, by some miracle, a significant boycott of AI occurs, the infrastructure that has already been built and what is currently in development won’t be dismantled due to lack of use. Instead, the industry will invent new uses because there is no idle capacity, and more importantly, no one is investing an obscene amount of money just to see if it will pan out.
One thing is a handful of startups selling empty promises backed by almost-cartoonish villains like Marc Andreessen, as we saw during the NFT craze. Another is the largest companies and investment funds on the planet pouring hundreds of billions of dollars (literally) into a technology they believe to be revolutionary.
When it reaches this stage, there are only two possible outcomes: a massive return accompanied by a true revolution, or widespread failure, regardless of who it affects. You can probably guess my hunch about which outcome we will see.
***
It’s a bit frustrating when the responses to a dilemma echo the broad and, at this point, tired phrases like “it’s capitalism” or “these are systemic problems, we can’t do anything as individuals.” Even if that’s the case. So, I want to wrap up by providing a bit more context, on an individual level, to help ease any guilt you might feel if you want to use ChatGPT for any tedious tasks you’re asked to do at work or, let’s be realistic, to experiment with the available tools — who knows, you might find something useful?
You know those video calls, Zoom and FaceTime, that saved some semblance of sanity during the pandemic and normalized video chatting? They are among the most energy-consuming digital activities. And it doesn’t have to be that way if we’re willing to turn off the video and just use audio, like we did before the pandemic. This simple gesture saves 96% of the carbon emissions from a video call — or, in this case, a call.
Before you delete the Zoom app from your phone, consider that, even if you don’t turn off the video for any reason, a video call is obviously much more economical and energy-efficient than flying and then taking an Uber or renting a car to have a face-to-face conversation with someone on the other side of the country.
You know what consumes way more energy than hundreds or thousands of silly responses and bad code generated by ChatGPT? Video streaming. Are we all going to cancel Netflix and YouTube and go back to the 1990s internet, which was all text-based, and to DVD rental stores? Yeah, I, along with Blockbuster, don’t think so.
For my part, I believe that directing (our!) energy toward actors, instances, and problems with a higher chance of yielding greater (systemic!) returns is a more promising path. How? I don’t know. I leave that question (and my support) to activists, scientists, politicians committed to the environment, and anyone else who knows better to help. I’m just an average guy writing on a blog that almost no one reads.
Let’s fight against the establishment of data centers in areas already suffering from drought, or push big tech companies to be more transparent in their environmental reports and more efficient in water reuse and developing techniques to mitigate environmental impact.
Regardless of what we think about AIs in general, and generative AIs in particular — whether they’re tacky, unethical, super cool, or revolutionary —, I know that harassing those who use them won’t solve the problem. I mean, for the use itself. Making fun of someone who claims to be the author of poems generated by ChatGPT, well, I think that’s fair game.
28/1/2025
Google will change the names of the Gulf of Mexico to the Gulf of America and revert Denali to Mount McKinley in its Maps app, aligning it with one of President Donald Trump’s weird ideas.
Someone dug up a 2008 post from Google’s public policy blog where the then-global director of the sector talks about this very issue — “How Google determines the names for bodies of water in Google Earth.”
Google has a uniform policy they call “Primary Local Usage:”
Under this policy, the English Google Earth client displays the primary, common, local name(s) given to a body of water by the sovereign nations that border it. If all bordering countries agree on the name, then the common single name is displayed (e.g. “Caribbean Sea” in English, “Mar Caribe” in Spanish, etc.). But if different countries dispute the proper name for a body of water, our policy is to display both names, with each label placed closer to the country or countries that use it.
In other languages, Google uses the common name in the language that Google Maps/Earth is being displayed in, along with an expandable button that lets you know the name isn’t universally agreed upon and lists other names that are also used.
That’s where people are giving Google a hard time, as they have (or used to have?) a policy that adopted the criteria of “primary, common, local” names for bodies of water:
[…] By saying “common”, we mean to include names which are in widespread daily use, rather than giving immediate recognition to any arbitrary governmental re-naming. In other words, if a ruler announced that henceforth the Pacific Ocean would be named after her mother, we would not add that placemark unless and until the name came into common usage.
On X, the company responded to the criticism by saying they have “a longstanding practice of applying name changes when they have been updated in official government sources.”
In the case of the US, that would be the Geographic Names Information System (GNIS). Note that the two changes — the Gulf of Mexico and Denali — haven’t been published by the GNIS yet, so Google Maps is still showing the “old” names there.
I get the frustration with the arbitrary decisions of an erratic president, but this seems like a… non-issue? If the government changes the names of bodies of water, as the US government plans to do (the executive order was published by Trump on January 20th and is pending a GNIS update), Google is right to reflect that in Maps.
