I’m writing this on an airplane on my way to spend time with family, which I’ve been looking forward to since I started my break a few months ago. I don’t have any retrospectives for 2024… I’m much more consumed by looking ahead to 2025. Development is changing with the introduction of all of these LLM-powered developer productivity tools. Digging into which ones are good and what problems are most important for my engineering teams - that’s one of the big puzzles for next year.
For now, here’s the normal list of interesting links for the last week. It’s a quick one. I hope you all have some time off to recharge and connect with family over this end-of-year holiday period. Happy Holidays and Merry Christmas to those that celebrate!
Reads #
-
The Developers Who Came in From the Cold: DaringFireball covered this, so I imagine anyone reading this has seen this already, but it was a great read about the challenges Rogue Ameoba had maintaining Audio Hijack, a critical audio app for many professionals, through various MacOS updates. Some nice lessons in there for working with platform owners in there.
-
The Ghosts in the Machine: I’ve never been a Spotify user. I prefer Apple Music. A tiny part of that has been the sense that Spotify doesn’t really care about music. I don’t think that’s entirely fair, but it never seemed like they really had solved the business model questions that still loom over the industry. This article reinforces that feeling. Spotify’s solution for trying to free music (and thus their profit margins) from the overhead of the labels is to just… commission music to fill their top ambient playlists. Good report by Harpers.
-
‘We Can Bury Anyone’: Inside a Hollywood Smear Machine: Not really a tech story, but interesting in that it describes some aspects of how crisis PR firms astroturf to shape public opinion. The confidence of the PR firm in this story is telling, and I’m curious how much the platforms themselves are aware or care about this sort of behavior.
-
Scaling Test Time Computer with Open Models: Think we’ll see this technique more often:
Rather than relying on ever-larger pretraining budgets, test-time methods use dynamic inference strategies that allow models to “think longer” on harder problems. A prominent example is OpenAI’s o1 model, which shows consistent improvement on difficult math problems as one increases the amount of test-time compute.
Code & Tools #
-
Meta Video Seal: Meta AI released a new video watermarking model that aims to embed watermarks in video that are imperceptible to humans and can survive many common video modifications/ edits. They posted a nice little intro video to their threads account. Looks very interesting.
-
TRMNL: Covered on this week’s Accidental Tech Podcast, and thus promptly selling out, TRMNL is an E-Ink dashboard with a plugin ecosystem and SDK. I love gadgets like this and have even built a similar device myself. They do support DIY boards, so I may be able to kill my custom ESPHome + HomeAssistant code and use their framework instead. The plugin ecosystem looks really good at first glance. I’ve ordered one and have grabbed the SDK while I wait for my device to come in next month.
-
Bridgy Fed: I’ve been playing with this over the last week. The project aims to connect the various federated social networks to help users take a little more control over the social graph. This spawned a new organization, A New Social. Seems like an interesting project. I’ve bridged this site into the fediverse and BlueSky. I’m actively setting this up, so we’ll see what happens when this post goes live today. Semi-related, I don’t want to run my own Mastodon server anymore, and have killed the instance I ran at tranquil.cloud.