I'm no photographer, but still tried to catch the blood moon last night (3am JST). I love eclipses. I really want to see a full solar eclipse. Focusing was so difficult. Night photography is no joke. Especially as the brightness of the moon keeps changing as the eclipse progresses.

Redeem the time, for the days are evil

Listening to music from the late 90s/ early 00s is so crazily nostalgic..

How I wish I could afford a bigger GPU... 12GiB is just not enough

This veelogging trend is making me question my aim to stay pseudonymous here. I know that this identity is not even perfectly protected so why bother. I’d rather actually just live life and let people see me. So much is gained by sharing with others. Seeing a person really adds to the human connection.

#asknostr What’s a good go-to “what is Nostr?” link to share to pre-nostriches? Video or blog would do. Simple is probably better. But would be nice to have one example for non-technically minded folk and one for the technically knowledgeable folk.

“We’ve got dust on our bibles ; brand new iPhones. And wonder why we feel this way” Just heard this in a random song called Dusty Bibles by Josiah Queen That is such a good line.

Where are the “AI is boiling the oceans” headlines ?

I was using @Fountain on my iPhone recently mostly. But then wanted to use it on my GOS phone for the first time in a while. No matter how many times I restarted the app and relogged in, it kept showing an old queue. So I gave up and went back to iOS for a few days. Suddenly today I opened my iOS fountain and it’s playing a very old version of my queue. I wonder if it somehow got updated from my GOS phone Sad :( Just clearing my very long queue and will fill it again over time. I’m kinda glad as it was out of control

I made a Metrics Proxy for monitoring self-served LLMs (via Ollama, vLLM etc.). Helps with just giving some metrics around how the LLMs are being used and how they're performing. If it helps anyone, try it out! View article →

The really sucky thing about using AI is that you can’t verify what is giving you responses. You cannot trust it. They can tell you the base model, but you have to believe them; and even if they could prove it, most #LLM providers are running orchestration layers on top of the #AI feeding it other data and allowing the AI to fetch its own (by proxy) external data. So there’s no telling where it builds its response from. The safest would be to self host. You cannot trust it verify the base models you have. And then you can run your own tooling/orchestration on the model. For now we don’t have better solution, so we just need to be cautious. But that seems like the only sane and safe future for me. Let’s make it happen

Ah I remember the old days as a kid. I remember when I was about 11 maybe 12, Windows allowed you to set an html page as your background and I remember loading all the computers at my junior school to show funny messages with “alert” on login. I also made a little C++ program to pop up a prompt for the user to “re-enter their password” and it saved it to my user directory in a txt file. In early uni I remember using wireshark to sniff traffic the then popular mobile messenger app from a J2ME emulator, and then wrote my own client that allowed contacts to use to acces forums, read weather and chat in groups etc. I definitely had a more explorative adventurous punk mind then… I miss that. I also don’t have time for that now.. this is a great episode

Fountain

Darknet Diaries • 155: Kingpin • Listen on Fountain

In this episode, we delve into the multifaceted career of Joe Grand, also known as “Kingpin.” A renowned hardware hacker and computer engineer,...