Colin Devroe on Tumblr

Christoph Rauscher, in a post about his 6th year running a weeklong workshop called Writing = Design (which sounds lovely):

It was so peaceful, this one week of human writing. We developed sentences from words, focussed on first drafts, and provided thoughtful feedback for each other. Our group spanned from 24-year-old students to 60-year-old design professionals, and not for a single thing in the world would I want to exchange those perspectives with some artificial machine-made fabric. I’ll try to take that spirit into my everyday life, aiming to de-pollute (is that even a word?!) my thoughts and words from the possessive narrative that “AI tools” are supposed to dominate my life. They don’t. And that’s called agency—so glad I still have that.

Merlin Rebrović:

… I’m using AI to design AI-powered tools and I work with some people who are AI-pilled. When I get home, I read and listen to AI maximalists and billionaires’ manifestos so that I know what my executives are consuming and to get ready for what I’ll hear from them at work next week. I also follow people on the opposite spectrum, who talk about societal, environmental, and economic harms of AI, as well as predatory marketing tactics and lobbying employed by big research labs. Wading through all of that continuously takes energy and time.

I’ve been working in tech since 1994. Tech has always moved quickly and I’ve always done my best to jog along and keep up. Over the course of these decades there have been several times I’ve felt the need to sprint.

To me, a sprint is not sustainable. It is a short period of time where I must almost totally consume myself in order to do my job well. Sometimes the sprints are necessary because I fell too far behind and I needed to catch up, other times it was because I wanted to be at the head of the pack. Most of them have been relatively short-lived sprints; lasting perhaps half a year or so.

This AI sprint seems to only be getting faster and lasting longer than any other sprint I’ve experienced.

It can be argued when the sprint began. Artificial Intelligence has a much longer history than just the last few years. In fact, much of the work in the field was done before I was born. But this most recent race is the one I’m personally taking part in.

As Head of AI & Product at NerdPress, I’m obligated to keep up with what is happening in the industry, what the tools are capable of, and how best to apply them to our team’s work. It is a lot. And I feel Merlin’s words above where it is a bit consuming. But I took it on. I decided to start sprinting. I wasn’t asked to. Which I’m sure helps to make it a bit easier for me than others who were forced into the sprint.

Will it end? While I’m a little bit tired, a little frazzled, from nearly a year of all out sprinting, I think the pace is about to quicken yet again.

But now I have help. I’m no longer running alone. I have several AI agents to help me. To some, that sentence feels icky. But I’m totally OK with it. Even as the pace continues to quicken, I feel empowered to keep up much easier today than I was able to even a year ago. And I think the next year or more is going to be at least three times more work to do than the previous year.

Tyler Hall:

The first version, from November 2020, was called AntiPhoto. The name was a mood. I had tens of thousands of photos and videos scattered across drives and old phone backups, and Apple Photos wanted me to live inside its library, on its terms. I didn’t want a walled garden. I wanted something that could point at a messy folder and just make sense of it.

I linked to Tyler’s project over 5 years ago. I was excited about any new Mac-based solution for photo management at the time. And I’ve tested several builds of Iris Photos since (thanks Tyler).

The 1.0 release looks great. Many of the things the app does out of the box I do within my Apple Photos library (such as have smart albums for dates and people).

The only downside (for me) about Iris Photos is that it prefers that your entire photo library exists locally… preferably directly on your Mac. I simply do not have the disk space for that. So since my reluctant switch to iCloud Photo Library I’ve been using that exclusively.

Models. Big huge scary models. Small, purpose-built models. Slow models. Fast models. Expensive models. Local models.

Models. Models. Models.

Which model should you pick for your question, task, or goal? How much of your usage budget will you use when you do? Could it have been done by a lesser model for less usage?

As the frothy sea of AI continues to ebb and flow this seems to be one of the more challenging and hard to answer questions for those of us using AI agents to help us with our everyday workload.

The answer to this question doesn’t yet seem clear.

Orchestration; which asks a single agent/model pair to delegate work to lesser agent/model pairs is one leading answer. And it works in some contexts. For example: “Do this big task” can be planned with a larger more capable model and then delegated to one or several more sub-agents and cheaper models to do laser targeted work. In practice, this works pretty well and the tooling for this is beginning to mature to the point it happens invisibly.

But, for the most part, the leading orchestrators use their own cloud-based models (and usually only a very small subset of them). Meaning, Claude might spin up 10 sub-agents on Sonnet to do the work, but you’re limited by still needing to use an Anthropic-based cloud model.

Something I played with a fair deal is using prompt heuristics to determine which model may best suit the current task. I created a catch-all “model” that delegated to other local or cloud-based models based on the request, whether or not it included private data, etc. It works fairly well but there are so many model variants that keeping up with whatever is the latest (that can run on my hardware) is nearly a full-time job… and getting an agent to delegate the tasks rather than just single-threaded prompting isn’t easy. And isn’t something I’ve considered cracked, yet.

But I think we’ll see the solution to this question appear within the next year. And I am sort of hoping it will be in an unexpected way.

Just look at the options we have in front of us.

Take, for example, something incredible like chat jimmy. chat jimmy runs an older, less capable, Llama model but runs inference near instantly. Seriously, give it a try. It is so fast it almost doesn’t matter that it isn’t nearly as good as more capable models. Think of this; if chat jimmy needed to run 10 times to produce valuable results, it would still be 100x faster than today’s leading models.

Now, imagine if Apple, AMD, or Intel were able to take a somewhat more capable model and grind it down into the bits necessary to put directly on the silicon they ship with their products. And imagine if all agents could utilize these on-device super fast models for some tasks and only rely on the larger cloud-based models for planning and orchestration and more important things like security and testing?

I think we’re on the cusp of something like this being available. I’m already using something like it locally — a main, orchestrating agent and super fast sub-agents that only do very targeted work. But I don’t think it is the solution just yet.

As the frontier models grow, the local models are shrinking. As the frontier models chase general intelligence, the local models are focusing on very specific sets of tasks.

I think overall the gap between frontier cloud-based models and local models will only widen and, eventually, for most of our everyday work we’ll be able to use a suite of smaller models that are orchestrated by agents. The frontier models will try to solve protein folding while our local models will write website code.

It could be argued that Apple believes this already. What they call Foundation models are really just a set of models that are purpose-built for the task they are supposed to do; image recognition, speech to text transcription, emoji generation. Everything else they leave up to the frontier models.

Famously, for years I have read every single blog post in NetNewswire, every tweet on Twitter (when that website existed), and every toot on Mastodon for the websites and accounts that I follow. It was sort of a disease that I couldn’t shake. I felt like I had to read every single one.

That is, until recently.

After a long week away I found myself struggling to catch up. And I tried. I really value the insight and experiences of all the people and accounts that I follow across the web. I’ve learned so much from them. But if it stops being fun or valuable and causes me stress I thought it would be good to reevaluate my process.

I’ve taken social media breaks several times. And when I do, I declare a bit of bankruptcy rather than trying to catch up after a month or more away. Each social media break I’ve taken I found worthwhile.

Recently, though, I’ve just given up completely on the idea that I can catch up. So I think I’m over the habit of even trying to read every single post.

For the last several months, I’ve not been keeping up on reading RSS feeds or Mastodon. I fall far behind and then quickly skim the entire timeline to pick out things to read. But even that practice I’m starting to eliminate.

In NetNewswire I’ve refactored my previous setup. I used to keep folders of feeds called “Trialing” and “Subscribed”; if a feed was valuable enough for a long enough period of time it would be moved from “Trialing” into the “Subscribed” folder. Or it would be deleted. These days, I changed to a “Some posts” and “Every post” folder. The “Every post” folder is very, very small selection of feeds. So for all feeds I either read some posts or every post. And I’m slowly whittling those away.

I’ve completely given up trying to read every toot. So now I use the official Mastodon iOS application or the website (I’m no longer subscribed to Ivory as development of that app has seemed to stagnate) and I open it occasionally and read the most recent few posts and then move on with my day.

I have no desire at all to use any apps that combine multiple network feeds together. As Dave Winer would constantly remind everyone; that is what RSS is for! We don’t need all these apps! But, no one listens.

Tracy Durnell is shaking things up:

I’ve done a purge of my feed reader (down to 80 feeds lol). Rather than defaulting to my phone, I’m recommitting to books as my after-lunch reading.

She goes on to list some new rules to help decide what to read online.

Readers of my blog may recall, I do this routinely. I gobble up everything for a time, then I prune (more like lopping) way way back. Then I start over.

The AI slop explosion is real and I’ve not dodged it. In fact, I’m right in the thick of it. It has been a sprint to keep up with. But I feel like my reading Tracy’s post, albeit a few months later since I’ve been so busy I’ve neglected my feed reader, is as good as any reminder to scale back and revisit my information intake yet again.

There is no white-walled room, pre-recorded, distillation of the essence of chamfered edges here… or talk about liquid anything. This post is about how Signboard is quickly becoming so feature-rich you likely won’t use them all.

Signboard 1.5 just shipped and the landing page has been updated.

This release brings much tighter Obsidian integration (including a plugin which will be released into the community there soon) with linked notes, full folder conversion and support for Base file generation. What does all of this mean? That placing your Signboard board folder into your Obsidian Vault makes it super powerful to use the two apps side-by-side.

Here’s a Signboard + Obsidian use case; these days we’re recording and transcribing everything. Zoom and phone calls. We’re pulling down entire websites into Markdown. Allowing agents to run amok all over the web gathering research for us. Where do you put all of this data? Everyone is using Obsidian for this of course. Signboard isn’t the right place. Signboard is where you track what you need to get done, what you’re doing, and what you’ve completed. But now you can link a card with an Obsidian note or folder.

A linked Obsidian note in Signboard 1.5

This is far more useful than it may seem. And agents know how to use it too. So now your Signboard cards, which can link other things like images and videos and documents, become your one-stop shop for everything you need to get your work done.

This release of Signboard also brings a more dense table view (in addition to the already powerful calendar, this week, day, and agenda views in the planner area).

The new sortable table view

Signboard 1.5 also extends the number of boards you can have open at once from 6 boards to unlimited boards. I find the board switcher (CMD + K) to be critical to switching between my projects. But, what I love is that the planner view brings them all together in one place – no matter how many boards I have open in Signboard.

You can also keep your system calendar up-to-date with any tasks or cards that have due dates in your Signboards by publishing a local calendar feed.

The macOS calendar with Signboard tasks