It’s 2025. Let’s stop grading generative AI on a curve

Compared to other tech giants, Apple’s approach to generative AI is surprisingly measured. Its new “Apple Intelligence” errs on the side of error-prone or exploitable functionality: Image Playground, for example, is nothing that could be mistaken for a photo, but only produces cheerfully synthetic Pixar-esque graphics.

But even Apple can do a lot to reduce AI’s tendency to go rogue. Another Apple Intelligence feature uses technology to summarize notifications. Last month, details of a BBC article about Luigi Mangione, who was charged with killing United Healthcare CEO Brian Thompson, mistakenly stated that Mangione had fired the shots. himself. Obviously, the BBC was not happy about this. Reacting to both this botched summary and an earlier one claiming the arrest of Israeli Prime Minister Benjamin Netanyahu, a group of journalists said Apple should simply abandon notification summaries.

It’s a dystopia for all companies to have a new AI feature from Apple spreading misinformation about the world’s most watched news stories. And that’s before we get to the fact that Apple Intelligence summaries, even if accurate, often read like they were written by an alien with only a vague familiarity with the way people live. Given that they primarily compress items that are quite short, it’s not clear that summaries are a net positive for users of Apple’s platforms. I can’t imagine the company will kill this feature, but maybe they should have held off on releasing it until it worked better.

The race to implement AI in the tech business has led companies to ship functionality that they know is crude and unstable—not because it’s wrong in any traditional sense, but because the system is unpredictable. For an industry used to working in ones and zeros — making quality control literally binary — this is a sea change. Thirty years ago, it was discovered that there was a bug in Intel’s Pentium processor that could cause numbers to be divided incorrectly. The chance of a flaw affecting any calculation was one in nine billion, which initially led Intel to downplay the flaw. The chip maker apologized after realizing its happiness was a bad look and spent millions to provide tuned Pentiums to PC owners who demanded them.

With today’s AI, flaking is so central to the experience that some of my conversations with Anthropic’s Claude are dominated by self-aware acknowledgments of its limitations. In response to a recent query, he helpfully explained: “I should note that while I try to be precise, I can dream up the details when answering questions about very obscure subjects like this.” It’s better than the maniacally overconfident personas of other chatbots, but it’s still a break, not a solution.

It’s true that current versions of Claude, ChatGPT, and their competitors are far less prone to getting ridiculously wrong than their predecessors. But they remain very good at weaving plausible-sounding inaccuracies into mostly authentic material. Computer graphics have long suffered from an uncanny valley problem, where realistic cartoons that are 90% believable tend to make viewers root for the 10% that are imperfect, such as eyes that lack a believable human glow. Artificial intelligence, on the other hand, has a valley – a percentage of the material it generates is wrong, but clearly not. This is a much more dangerous issue than the easy-to-detect hallucinations, and it’s not going away anytime soon.

Now I know I use this newsletter a lot about the flaws of AI. I swear I’m not a Luddite. I have the ability to dazzle with artificially infused features, and I don’t think they need to be perfected to be extraordinarily useful. For example, this week’s newsletter makes a formatting adjustment that I can’t figure out myself. Claude handled most of the coding with only general instructions from me. Even considering the time it took me to complete it, it felt like a miracle. (I’ve tried ChatGPT before and found his advice useless.)

As tempting as it is to carve out an infinite amount of space for something as mind-boggling as generative artificial intelligence, the best computing breakthroughs require no special dispensation. Arthur C. Clarke famously said that any sufficiently advanced technology is indistinguishable from magic. With all due respect, the opposite may be true: We know that a new technology is advanced enough when it is so reliable that we consider it mundane rather than magical. When was the last time you were in awe of the internet? Or electric or air travel?

If 2025 is the year that the novelty of generative AI wears off, it could force tech companies to hold themselves to the same standards as all the older, more familiar technologies at their disposal. That coming of age can’t come soon enough—and in its own way, it would be a giant leap forward.

Read / Watch / Listen / Try

At Fast CompanyJared Newman’s best new apps of the year — both new ones and meaty upgrades — are a holiday tradition. Here is his list for 2024. This has inspired me to recommend three additional productivity apps. None of them are very famous, but they are as important to my work as any of the big names.

bear Since ditching Evernote in 2023, I’ve tried countless note-taking apps. Available on Macs, iPhones, and iPads, Bear is the closest I’ve come to thinking like I do. Its interface is beautifully minimalistic, and I’m on board with its use of hashtags as an organizational tool. Even the Pro version is an absurdly reasonable $30 a year.

Reclaim.ai. This web-based app helps me accomplish tasks on my to-do list by intelligently placing them in an accessible space on my Google Calendar where it’s impossible to ignore them. There are paid plans, but the free version offers everything I need and more. Until now, I somehow missed that Reclaim was acquired by Dropbox in August; Here’s hoping this company doesn’t mess with a good thing too much.

Focus. In 2015, I learned – a Fast Company article, of course!—about a productivity technique called Pomodoro. This involves breaking your workday into 25-minute chunks, with short breaks in between when you need them. I continue to find it a boon to my efficiency, and my favorite way to manage it is with this nifty timer app for Mac, iPhone, iPad, and Apple Watch—not to be confused with Focus’s suite of other timers.

You read it The Plugged, Fast Company‘s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague referred you to this issue, or if you’re reading it on FastCompany.com, you can check back issues every Wednesday morning and sign up to get it yourself. I’d love to hear from you: email hmccracken@fastcompany.com with your feedback and ideas for future newsletters. I’d love to hear from you about Bluesky, Mastodon or Threads.

It’s 2025. Let’s stop grading generative AI on a curve

Read / Watch / Listen / Try

More high-tech stories from Fast Company

Leave a Comment Cancel reply