Gemini* and Gopher

• Personal/Opinions

  • 🔤SpellBinding: GHILUSO Wordo: CRISP

  • Supernatural Mystery

    I mentioned in my last post that a friend of mine and I are working on a supernatural mystery story. We have a good start, over 15,000 words. But it won't be ready for a long time. Below are a few of scenes from Chapter 2. We're going to go over them a few times, I'm sure, so this is just a rough draft.

    I chose these scenes since Lantashi (me!) is introduced. In this story I am just a sassy human, but a sassy human in the Upper Peninsula of Michigan, which is almost as good as being a Sylvan!

  • The paradox lies on infinity

    Back in December 22nd, I published the post "The endless desert at the afterlife" (I'm not sure how to link, but you can see it at the "Latest posts").

    Perhaps a bit of thought experiment, applied math and science are interesting to complement philosophy.

    Suppose a scenario similar to Conway's Angel Problem: two figures are roaming through an infinite plane.

    However, there are plenty of differences from the original thought experiment.

    First of all, no figure is trying to block the other's path, nor is trying to evade the other.

    They start at very different coordinates within the infinite plane and they both have a common goal: to meet each other.

  • [music] 2024-12-28

    GONE/AWAY is a different MAVS than i am used to, making use of field recordings out in nature to give a transportive effect, with beats only coming in briefly halfway thru the album. like a lot of ambient music i don't think i would ever be compelled to listen to it again, but it was a pleasant walk thru the forest & along the beach.

    much as i could grumble about spoken-word jankiness, once the Lifeless Dark LP got going with music there was super kinetic & rifftacular thrashing stenchcore to be had. the vocal production seems a little thin & clean to me sometimes but that's a nitpick on a solid debut that made me yearn for the pit. i bet these folks crush live.

  • Mechanical People

    Have you ever decided to help someone or present something, which eventually resulted in bigger damage than you expected? If so, what conclusions did you make from such situations? One of them was probably that there had been something wrong with the knowledge or experience you had at that time. So, the problem was with you. The other conclusion could be that something was wrong with the partner or partners you wanted to show, explain, and prove something. Maybe they were too bold towards you?

  • Mechanical People -> Devilish Implications

    Who are the ones who take advantage of these mechanisms/occurrences/behavior of mechanical people? What do they look like? Is there some sort of secret protocol that enables me to communicate with them? What language do they use? And how should I refer to them?

  • “Good as new”

    One day, my mother came home after gathering a vast collection of snakes, insects, worms, and spiders, from eggs and larvae all the way to rotting husks.

    She delicately arranged the maggots and other vermin inside my abdomen, her face held still in soft focus. I felt in that moment like she could love me.

• Technology and Free Software

  • Building new GGML backends for novel accelerators, how, challenge and opportunities (FOSDEM 2025 draft)

    GGML has emerged as one of the leading framework for efficient inference, particularly for large language modes. GGML is engineered with performance in mind and with great quantization support. More importantly. Its flexibility and growing community support make it an ideal platform for exploring new hardware acceleration options. Already some rare hardware (mostly China's alternative to western chips) have made it into GGML.

    My journey with GGML starts back in end of 2022 when LLMs are new, hacking GPT-3 is damn easy and RK3588 was new and exciting. Rockchip finally got around and made their matrix multiplication API working. And I was starting to get fairly ecoconscious. I was thinking, LLMs are power hungry running on GPUs. And I want a local AI assistant without a literal space heater in my room. Which in Taiwan can be 32°C indoors during summer! So I started looking into utilizing the RK3588 NPU in llama.cpp. It sorta works. But the NPU's design is really not suitable for LLMs and has miserable performance for such use case (worse then CPU!).