patwolf

Personally I like it more with the labels. Reminds me of web design where sometimes you have to add some unnecessary text or icons or else it looks too bare.

It's funny seeing "The Dixie Chicks" listed under "The Chicks", given they didn't change their name until 2020. I get it, but it's amazing how thorough they were retroactively updating content.

I had a Withings watch and absolutely loved it. I considered buying another one after my old one got water damage. However, it was becoming evident from changes to the app that it was getting more and more locked down and ultimately was going to require a subscription for anything useful. At the same time, the app was getting buggier. At one point I lost multi years worth of data when I had to re-sync my watch to a new phone.

It's a shame because I liked their spin on a hybrid smart watch which looked like a traditional analog watch. Garmin does make hybrids, but to me they all look too sporty or tactical.

I was hoping for something more novel. There must be dozens of companies out there working on rethinking what repos and workflows look like in the agentic era. It feels like we're all riding around in horseless carriages but haven't yet figured out the ergonomics of a modern automobile.

I only have a few digits of pi memorized, so I had better luck using digits from fibonacci's sequence. I can derive them in my head, albeit slowly as the numbers get large.

I've been experimenting with AI powered DAWs. I finally settled on using Pi Durable to embed the prompting directly in the UI.

I've also found AI good at building synths. I now have a bunch of synths, samplers, and drum machines that my DAW can configure through MCP. For example, I can tell it to configure a 303, generate a 4-bar arp pattern, slowly increase the resonance across the pattern, and add it to the song.

One of the reasons I use OpenRouter is because they offer zero data retention. As far as I can tell, DeepSeek's own API doesn't support ZDR.

I grabbed one of these since I didn't have a hardware FM synth. While playing around with a web-based patch editor and seeing how well synths can run in a web, I started using AI to build out my own synths, and it's been a lot of fun. I've done FM, granular, samplers, 303 and 808 clones, and a bunch of others. So yeah, if you don't care about the experience of having a bespoke device, you could probably one-shot something much more sophisticated than this.

This is the same argument from when Sirius and XM radio merged. It didn't matter that they were monopolizing satellite radio because the real competition was terrestrial radio and internet radio.

I used to use Claude inside of a terminal within Webstorm until I realized I rarely looked at the code in the IDE. Instead I was creating draft PRs and reviewing the code in github. Now I use claude inside of herdr, and I appreciate having more screen real estate devoted to claude. As a long-time Jetbrains subscriber it pains me to stop using it. But at this point, I'm not sure what they could come up with that would be compelling enough for me to switch back.

I've been using the web search in OpenRouter, which is similar in that it's a wrapper around other search engine providers. It's really convenient to be able to experiment with new models and new search engines without having to go through corporate hoops to subscribe to a new service.

I've worked with developers that were reduce maximalist. During PR reviews, anything that could be rewritten with reduce was flagged. One of the benefits of AI is not having to care as much about things like that.

IBM's J9 had AOT for a long time too. I wonder how JEP 544 compares.

I played around with codegraph. I did a quick experiment where I gave claude code a feature to implement, and I had it perform the same task with and without codegraph. This wasn't scientific, so take it all with a grain of salt.

The codegraph implementation used fewer tokens. However, I performed an automated code review on each implementation, and there were more issues with the codegraph version. These weren't major issues, more like it re-implemented a function that already existed, or used styling that wasn't consistent with the rest of the codebase. My theory is that there's useful information an LLM gleans while grepping through code that is missed when it uses codegraph instead.

Often in software development I encounter code that seems overly complex. It's not always obvious at first glance whether that complexity is because it's solving a really difficult problem, or because the developers made it more complex than it needed to be.

Having no domain knowledge of the problems of detecting cracks in blades, this sounds like either a genius solution to a super hard problem, or unnecessary complexity.