Maybe it’s my age showing but I’m with Gruber on this one:
I would argue, strenuously, that the phone is the natural AI device. It already has: always-on networking, cameras, a screen, microphones, and speakers. Everyone owns one and almost everyone takes theirs with them almost everywhere they go.
This isn’t entirely surprising but it’s a sad state of affairs, and it’s worth highlighting not just how, but also where LLMs are being trained:
Hundreds of thousands of hours of work goes into providing enough feedback to turn an LLM into a useful chatbot, and that means the large AI companies outsource the work to parts of the global south, where anglophonic knowledge workers are cheap to hire.
I know it’s too dismissive to call chatbots “fancy autocomplete” like many do, but we have to remember that this isn’t magic. The words the bots use come from somewhere. And in the case of “delve”…
I said “delve” was overused by ChatGPT compared to the internet at large. But there’s one part of the internet where “delve” is a much more common word: the African web. In Nigeria, “delve” is much more frequently used in business English than it is in England or the US. So the workers training their systems provided examples of input and output that used the same language, eventually ending up with an AI system that writes slightly like an African.
Some really interesting (and surprising) takeaways in this research, and a very good analysis by Caitlin Dewey in Actually, the internet’s always been this bad:
A team of Italian researchers evaluated more than half a billion comments spanning 30 years, and concluded that online discourse is no more ‘toxic’ today than it was in the early 1990s. […] Overall, the study found that the prevalence of both toxic speech and highly toxic users were extremely low. But the longer any conversation goes on, on virtually any platform, the more toxic it becomes.
Alex Heath has a really interesting interview with Figma’s CEO Dylan Field, covering life at Figma after regulators forced Adobe to abandon its $20 billion acquisition of his company. It covers a wide range of topics, but I wanted to highlight Field’s thoughts on generative AI, which largely matches my own viewpoint:
If I was to zoom out even further to knowledge work, we’re very much in a paradigm of AI as a tool and AI helping people get work done, but it’s not necessarily a replacement. I really think that there’s a human in the loop going forward in that AI might be a useful tool, but we all know its limits in terms of hallucinations, in terms of potential inaccuracies. Even if you apply it to rote tasks, it’s important to check the work. And you know better than anyone as a writer that the current models do not match your ability to write, let alone gain context in a conversation to ask the right questions or show the intelligence that you have as a journalist.
If you think about what it takes to create great design, there’s so much in that context window that’s emotional or thinking temporally about a brand experience or a user flow. I just don’t see how, in the near term, AI is able to have that as part of its context, which means that humans are providing that.
I love stories like this. Turns out there’s a sort-of movement of music being released on floppy disk… I will have to watch it from afar with admiration though. I’m already collecting vinyl and CDs so this is probably not a good idea for me.
There are almost 2,300 floppy releases listed on Discogs.com, most of which are electronic, but other genres include hip-hop, a smattering of classical and jazz, a bunch of metal subgenres, and “non-music” like experimental field recordings from Norway and spoken word from China. In 2018, Rolling Stone covered a “mini-boom” of vaporwave releases on floppies, noting that the lo-fi, lobit nature of vaporwave was an obvious match for the storage constraints of the 3.5-inch.
Here’s Lorin Hochstein with another great post about the practice of learning from software incidents. He asks, What if everybody did everything right?
An alternative lens for making sense of an incident is to ask the question “how did this incident happen, assuming that everybody did everything right?” In other words, assume that everybody whose actions contributed to the incident made the best possible decision based on the information they had, and the constraints and incentives that were imposed upon them.
I’ve been seeing a lot of this type of sentiment about React recently…
By my reckoning, if you’ve maintained a React codebase for the past decade, you’ve re-written your application at least three times and possibly four. […]
By choosing React, we’ve signed up for a lot of unplanned work. Think of the value we could have produced for our users and company if we weren’t subject to the whims of whatever the cool kids were doing over in React.
This is such a good point about music discovery and the abundance of choice:
Before Spotify, when presented with a new album, we would ask: why listen to this? After Spotify, we asked: why not?
I also like this sentiment:
On one level it’s impressive that Spotify can perfectly capture my musical taste in a series of data points, and regurgitate it to me in a series of weekly playlists. But as good as it has gotten, I can’t remember the last time it pointed me to something I never expected I would like, but ultimately fell totally in love with.
For that you needed someone who could go beyond the data to tell you the story: of the artist, of the genre, of the music they made. For that you needed criticism.
I’ve definitely seen this too! How ✨ became the unofficial AI emoji:
In the relatively short history of emoji, sparkles have been used to express excitement and magic, said Jane Solomon, the senior emoji lexicographer at the emoji reference site Emojipedia. Branding new AI products with the ✨ emoji suggests that these tools are exciting and magical, which might encourage more people to test out the technology. “It can seem like magic if you don’t understand how it works,” Solomon said.
My work at Jeli so far has given me a new lens on “incidents”—both in the software world and beyond—that I didn’t have before. These “failures” are everywhere around us. But are they really failures? Or are they ways for us to learn more about the systems we work within, and how to improve them? I think it’s the latter, and My $500M Mars Rover Mistake by Chris Lewicki is another story that showcases that…
The core lesson I’ve drawn from my rover ordeal is best expressed in these words: Let your scars serve you; they are an invaluable learning experience and investment in your capability and resilience.