/reader

Molly White has a post today about many of the good things AI can do, along with the costs and consequences of getting too wrapped up in the hype:

I’m glad that I took the time to experiment with AI tools, both because I understand them better and because I have found them to be useful in my day-to-day life. But even as someone who has used them and found them helpful, it’s remarkable to see the gap between what they can do and what their promoters promise they will someday be able to do.

Many of the most useful capabilities are actually not very exciting. As an example, for years Apple and Google have used machine learning to improve photo search, so you can find photos that include pets or buildings or concerts. Today, AI can take that kind of feature and super-charge it, with remarkably accurate photo summaries.

AI is not going to fizzle out like the blockchain. More and more software will embrace AI, in some cases going too far, sprinkling it throughout apps without any transparency into how it’s used. I could start to feel the temptation to go down that path too, which is why I took a step back to add the global AI setting this week, before we even have anything new that uses the setting.

No major feature should be so intertwined with AI that it can’t be clearly documented and controlled by the user. Humans are the ones who think, write blog posts, share photos, and join conversations. Some developers will push the limits, but not Micro.blog. AI will be a tool to help us, in narrow, practical ways, and I’m not interested in going much beyond that.

I like writing blog posts by hand, but I think AI can be a powerful tool. Micro.blog Premium currently has two AI-related features:

• When you host your podcast on Micro.blog, we’ll automatically generate a transcript of each episode, which you can edit.
• When you bookmark web pages in Micro.blog, we’ll automatically generate a short summary of the web page, which you can show on the web under the “…” button → Show Summaries.

As we add more features that are powered by AI, we need a comprehensive way to tie them together, especially for anyone concerned about privacy. To be clear, there are no plans to use AI except on public data that is already on the web, like podcasts or photos. We aren’t going to suddenly start sending your private Micro.blog notes to OpenAI for processing. (And because they are end-to-end encrypted, we couldn’t anyway.)

This is all a long introduction to announce a new, simple setting to globally control whether AI is used for anything in your account. You can find it on the web under the Account link:

For existing users, this is turned off by default unless you have previously used one of the AI features mentioned above. For example, if you have a podcast that generated transcripts, the setting is on so that the feature will continue to work. For new users who join Micro.blog starting today, it’s turned on by default. I think it’s simple and clear, but obviously we’ll be listening to feedback.

I recommend toggling the checkbox on for your account if you are at all interested in the potential for AI around things like helping with photo alt tags, search, or categorization. No new features to announce today, though.

This is not a lament for Humane or its business model. It’s a lament for all of us. So many innovative products will never get funded or never launch a product because if they can’t connect deeply with the smartphone, they’re at an impossible disadvantage.

Generative AI is such a unique technology that it is useful even completely outside the Apple and Google ecosystems, but some features just aren’t possible without open platforms.

Brian Chen at The New York Times was a little more positive, but still pointed out the many shortcomings:

I liked the chic aesthetic and concept of the pin. It was occasionally helpful, like when it suggested items to pack for my recent trip to Hawaii. But as I wore it for two weeks, it presented glaring flaws.

Ken Kocienda has a good blog post about working on the AI Pin, how no tech product is without trade-offs, and what things might be like in the future:

I think that LLMs are an advance on par with CPUs. Decades ago, the development of the first CPUs inspired people to make operating systems and programming models and, eventually, personal computers that allowed people to take advantage of the technology without being technologists themselves. We are at a similar moment now with LLMs, and the rate of change in Ai is far beyond what we saw decades ago with PCs. We built the Ai Pin with this historical perspective in mind.

There are interesting ideas here but it feels like NewtonOS 1.0 — just a little ahead of its time. If Human has the money to iterate, it could be something. My unsolicited advice: drastically simplify, scrap the laser projection, and focus on speed. A smaller, cheaper, AI “button” that was nothing more than ChatGPT available everywhere would be really fun to use.

I’m still excited about receiving my Rabbit R1, too. I was late in the pre-order queue, so likely won’t know what the device is like until much later this year. This is the time to experiment, and I don’t think companies like Humane or Rabbit should be judged too harshly on their first attempt. We’ll see where this goes.